Building the perfect Wildlife Camera (Guide) – What we learned and why we pivoted to Software

Thinking about building a wildlife camera? Or maybe you already have plans in motion? Either way, you’re in the right place!

At Animal Detect / Really A Robot , we built our own wildlife cameras, one with novel features. In this article, I’ll share what we worked on, the feedback we received from researchers and conservationists, and most importantly, why we ultimately stopped building cameras and shifted our focus to software.

How It All Started:

Really A Robot, where Animal Detect was born was founded by me and Eugene Galaxy during our Robotics Master’s at 丹麦奥尔堡大学 . (They love to boast about being the #1 Technical University in Europe and #4 in the world while we studied there, so of course, I have to mention it ??).

From the start, we knew we wanted to invent, build, and develop solutions that could have a real impact, whether in robotics, software, or hardware design. After working on a variety of exciting projects, we had several directions to choose from. So, what did we pick? Animals!

Both of us have always had a deep love for wildlife. Eugene had long dreamed of developing robotics to help marine life, while I have always been fascinated by observing animals in their natural habitat, whether it's a tiny squirrel or a massive moose.

With a mix of funding and support from our consultancy work, we scraped together a small but skilled team to bring the Animal Detect Camera to life. However, the idea wasn’t initially ours, it was our good friend Nikolaj Michelsen Kristiansen , who was involved in Danish wildlife conservation, which introduced us to real-world problems researchers were facing with existing camera solutions. He helped bridge the gap between us as engineers and the actual needs of conservationists and field researchers.

The quest for the perfect Wildlife Camera

Throughout our journey, we collected a ton of feedback, some of which we implemented, while other ideas were noted for future reference. My hope is that by sharing what we learned, anyone developing a wildlife camera can avoid some common pitfalls and benefit from the insights we gathered.

Now, what does the perfect wildlife camera look like?

Simple: It should do everything and cost almost nothing.

Of course, that’s easier said than done. Let’s start diving into the specifics of what “everything” really means, let’s look at some of the core features we built into Animal Detect and why they mattered.

Online (Cellular, GSM, 2G/3G/4G LTE) or offline (SD card) cameras?

Well, it depends where, who the user is and what the trade-off is (sounds super generic... I know). While we initially thought that using an online camera is a complete no brainer, especially where cellular data is available, we meet surprisingly many people who didn’t care about it. The constant notifications on a phone, real time alert and live camera functions are important for some, but not others. Some people deploy cameras for a period, collect them and analyze the data, before deploying the cameras on new spots. Additionally, with cellular cameras, there is an additional cost for subscription such as SIM cards and sometimes additional cost included in using and storing photos on servers. However, in some cases live updates and even the possibility to access the camera feed are needed, such as human/animal conflicts, poacher detection, capture and release scenarios and a bunch of other cases where getting live data is crucial.

Broadcasting and sending images over GSM also have other trade-offs, such as power consumption, possibilities of bad coverage and often the need to limit the resolution and size of images and videos, to reduce the data transmissions needed. So, who is your primary target group, and what do they need?

Don’t make assumptions, ask them!

While mentioning offline and online cameras, we also have a group of people wanting something completely different. Satellite connectivity! There are still several places where we need instant, or close to instant images, where there is no GSM coverage, with the expansion and progress of ways to transmit data over satellite, there are areas where online cameras previously were not able to send data, where with satellite connectivity there will be!

The actual “camera” image sensor(s):

Well, as good images as possible? A $10,000 DSLR camera quality, please! ?? While crisp images are nice and sometimes exactly what people want, we often heard something completely different: “I just want to be able to see the animal(s) in the image, that’s it.” Apart from the pixel count, there are quite a few other things to consider when choosing a camera sensor.

One of the early challenges we came across was that certain animals seem to stare directly at the camera at night when an image is taken. While infrared lights have some role in this, studies suggest that it’s often more than just the lights. Inside most traditional wildlife cameras, there is something called an infrared cut filter. What does this mean? Well, inside the camera, there is a photoelectric sensor that detects the light level, and when it gets dark, a mechanical switch moves an infrared filter in front of the lens. If you bring a wildlife camera into a dark room and turn the light on and off, you’ll often hear a “click” as it switches between day and night mode.

Here’s where it gets interesting. Some animals seem to hear that click or even pick up on small electrical noises from the camera. For those who have used these cameras, you’ll also know that this mechanical filter is one of the most common failure points. When it breaks, the camera gets stuck in either day or night mode, making it useless in some situations.

You can read a paper about it here: ?Camera Traps Can Be Heard and Seen by Animals

So how can we fix this? Well, we didn’t find the perfect solution, but one way to avoid the clicking issue entirely is to remove the moving parts. That’s what we did with Animal Detect cameras, where we used a NoIR (No Infrared Filter) camera, specifically the Raspberry Pi Camera Module 3 NoIR. This immediately solved two issues: no more clicking between day and night, and no mechanical part that could break.

But of course, this led to a new issue. A NoIR camera works both day and night, but since it captures a wider range of light wavelengths, the colors during the day can sometimes look strange. In some cases, the difference between a regular RGB camera and a NoIR camera is barely noticeable. But in other cases, especially when there are a lot of green plants, you’ll see bright pink leaves! This happens because some plants reflect infrared light, which a NoIR camera picks up.

While not ideal for beautiful nature photography, spotting animals was still no problem! If you want to read more about the Raspberry Pi Camera Module 3 NoIR, I highly recommend reading this article by Jean-Luc Aufranc: Raspberry Pi Camera Module 3 Review

So what’s the correct way to solve this? Well, the best way is to use a dual-camera setup. Instead of trying to make one camera work for both day and night, you can have one camera optimized for daylight and another specifically for night images. That way, you avoid mechanical failures, get proper color accuracy, and still capture high-quality infrared images at night. Here’s an example of a dual-camera setup:

Trigger Time

I’m not sure if this is the right place to talk about trigger time, since it depends on multiple parts of the camera, but let’s do it anyway.

What do we want? Near-zero seconds! But as long as we get below 0.5 seconds, we’re in the safe zone.

Trigger time is simply the time between when the motion sensor detects movement and when the camera actually takes a picture. Ideally, this should be instant, but also fast enough so the animal is still centered in the frame, not just a blurry tail at the edge of the image.

Trigger time depends on several factors, including the speed of the motion sensor, the camera’s shutter time, and how fast the system processes the signal. And yes, if a cheetah sprints past the camera, a fast trigger time would be essential to capture it properly. While some people don’t care too much about it, for those who do, the lower, the better.

The “Light” – IR Emitters and More

While we mostly rely on infrared (IR) emitters to illuminate wildlife scenes at night, there are cases where regular visible light, like a flashlight, is used instead. Regular lights provide direct illumination, but they often startle animals, making them less ideal for many wildlife applications. IR, on the other hand, is “invisible” to most animals, allowing for less disturbance while still capturing clear images.

That said, there are some situations where visible light is actually needed, though I’m still a bit puzzled about when and why. I’m no expert in this area, but having the option to switch between IR and LED modes on the same camera could be something worth considering.

How to Choose the Right IR Emitters

When it comes to IR illumination, the goal is simple: see as far as possible, illuminate the scene evenly, and avoid scaring away animals. But how do you achieve that?

One of the key factors is the wavelength spectrum of the IR emitters. Most animals (and humans) can’t see infrared, but even when an emitter is rated for a specific wavelength, that number only represents the peak, it still emits some light at lower and higher wavelengths.

For the Animal Detect camera, we tested multiple options and ultimately went with 950nm as it seemed to be the sweet spot. It was invisible to most animals while still providing sufficient illumination. Now, all that’s left is figuring out the rest, field of view, input voltage, and emitter placement, good luck with that! ??

To find the best setup, we tested a bunch of different IR emitters, including the OSRAM SFH 4544 - 940nm. While the peak emission was around 940nm, the actual wavelengths extended down to 850nm, meaning there was still some visible IR spill.

These emitters worked well but came with some downsides: They were expensive and they were through-hole mounted, making assembly more complicated

We went in another direction. After extensive testing, we switched to surface-mounted IR emitters from XSSY, as far as I know, we’re the only wildlife camera using surface mounted IR emittors, but I have no doubt we’ll start seeing more cameras follow this approach.

To optimize coverage, we combined 30-degree and 60-degree field-of-view (FOV) emitters, creating a more balanced illumination spread.

The camera window in front of the emitters acted as a diffuser, giving us an unexpectedly even light spread, not planned at all, but a happy accident!

But… The Images Are Still Dark?

Even with good IR emitters, night images can still turn out darker than expected. Fortunately, software can help with that!

One solution is CLAHE (Contrast Limited Adaptive Histogram Equalization), which enhances visibility in low-light images without overexposing certain areas. I’ve written about this before check out my post here: ?? CLAHE for Wildlife Images

And while you’re at it, blurry images are another challenge in wildlife cameras. I also covered how to reduce motion blur in wildlife images here: ?? Dealing with Blurry Images

Some IR Recommendations

For the best results, “No Glow” 950nm IR emitters with a black coating seem to be the way to go. They are significantly less visible in total darkness compared to clear-lens IR emitters, making them ideal for wildlife applications where stealth matters.

Fun Fact!

Can’t tell if an IR emitter is on or off? Just point your phone camera at it! Most smartphone cameras can detect infrared light, so you’ll see a faint glow on your screen even if it’s invisible to the naked eye.

The PIR Motion Sensor and How to Deal with False Activations

Just like everything else, we want to spot animals as far away as possible! But can we do that without false positives, like a branch moving in the wind?

Well, there’s a lot to say about motion sensing, but I’ll stick to Passive Infrared (PIR) sensors, since they are used in probably 99.9% of all wildlife cameras. The PIR sensor has one critical job, detecting whether an animal (or person) passes in front of the camera. It wakes up the system and triggers the camera to capture images or videos of the environment, and hopefully, some wildlife! ??

How PIR Sensors Work – And Their Weaknesses

If only it were that easy... A PIR sensor works on a simple principle, it detects heat changes between its sensing zones. If something moves and causes a heat difference, the sensor triggers an activation.

Sounds great, right? Well... what if it’s not an animal, but just a tree branch heated up by the sun and moving in the wind? PIR sensors can’t tell the difference, so you get a false activation.

Some sensors claim to address this issue with "white light immunity," which is supposedly a useful feature for outdoor PIR sensors to reduce false triggers from non-animal sources. We decided to use the LHI 968 sensor, which is also used in many other cameras that people love.

Things I wish I knew before...

One important hardware design tip: Place the PIR sensor 3-4 mm away from the PCB! Use a spacer or something to ensure proper placement.

Why? Because if the sensor is mounted too close, you’ll start hearing complaints like, "My other camera with the same sensor has way better detection range!" That small gap makes a huge difference in performance.

Making Sensitivity Control Smarter - Most wildlife cameras offer just three PIR sensitivity settings, low, medium, and high. We thought, that’s just not good enough!

So, what did we do? We added 1024 different sensitivity settings. Yes, 1024! Instead of pre-set sensitivity levels, we used a digital potentiometer (DigiPot) that allowed users to fine-tune the sensor sensitivity directly from a slider in our app.

Fighting false activations – Hardware level:

Now, how do we reduce false activations caused by heated branches, moving leaves, or other non-animal triggers?

We tried a lot of different approaches. The first attempt was a "blind" method, we assumed that if the camera started getting frequent activations in a short period, we could automatically lower the sensitivity.

Here’s how it worked:

1. If too many activations happen within a short time, the camera gradually reduces sensitivity to stop excessive triggers.
2. Once activations slow down, the camera gradually increases sensitivity again.
3. The user's selected sensitivity level would always act as the upper limit.

In practice, this actually worked! We significantly reduced empty images during the day when there were a lot of false activations.

But… it also introduced a new problem.

In some cases, a lot of real animal activity was happening, but the system interpreted it as false activations and started reducing sensitivity, which meant we missed real animals. So while it helped in some cases, it also backfired in others.

We kept refining the approach, but false activations are still one of the hardest problems to solve in PIR-based wildlife cameras.

Removal of Empty Images

I thought this needed its own section, as an extension of the "Fighting false activations – Hardware level:"

Alright, what about a not-so-blind way to spot empty images? Instead of simply reducing false activations at the sensor level, can we predict whether an image contains an animal without using AI? Surprisingly, yes! If you're working with limited processing power and want to process images directly on the device, this could be a viable approach. I tested out a non-AI image processing method, which you can read more about here: ?? Non-AI Image Processing for Wildlife Cameras

Using AI for Removing Empty Images

Of course, another approach is to use AI to filter out empty images. We experimented with this using a Raspberry Pi Zero to capture and upload images. While the RPi Zero is capable of running small TensorFlow Lite models, the size constraints mean the performance and accuracy are limited.

If you step up to more powerful hardware like the Raspberry Pi 4, Compute Modules, or Jetson Nano, you can run larger AI models directly on the device. But there’s a catch, power consumption becomes a serious issue. If you can figure out an efficient power management system, you open the door to running advanced models on the device itself, eliminating the need to send images to the cloud.

Our Solution? AI in the Cloud!

Ultimately, we found that running AI in the cloud was a better trade-off. While we didn’t reduce data transmission since all images were still being uploaded, we eliminated hardware limitations. This allowed us to use larger, more accurate models, significantly reducing the number of empty images being stored.

I briefly covered our empty image reduction results in this post: ?? Tech for Wildlife – MegaDetector & AI for Image Filtering

We also tested multiple AI models, experimenting with different classification strategies. One approach was not just filtering based on “animals” but also detecting features like whiskers, fur, and beaks. While it worked in some cases, it also produced more false classifications than expected.

The Risk of Removing “Empty” Images

While reducing empty images seems like a win, are we losing valuable data in the process? Some researchers and conservationists prefer keeping all images, as even “empty” shots might contain useful environmental data, subtle motion cues, or secondary research opportunities.

I wrote about the potential risks of over-filtering images here: ?? What You Might Be Losing by Removing Empty Images

Wow, that’s a lot of links to my previous posts… but hopefully, something here helps you build a better wildlife camera. Now, let’s continue! ??

The Fresnel Lens

As an extension of the PIR sensor, the Fresnel lens works in magical ways. You can think of it like a camera lens, depending on its shape and placement, you can extend detection range, improve accuracy, and focus infrared light in different ways.

The shape of the Fresnel lens, along with the distance between the PIR sensor and the lens, is all about math. Picking the right shape? Well, that part is actually quite simple, it should match the aspect ratio of your images.

For example, you’ll notice that the Boldygard MG-series cameras use a more square-shaped Fresnel lens, and their images have a square aspect ratio as well. Meanwhile, Spypoint cameras, which capture rectangular images, use a rectangular Fresnel lens. The placement of the Fresnel lens and PIR sensor is also critical, if the lens isn't positioned correctly, the camera might trigger for movement that isn’t even in the frame.

How the Fresnel Lens Works

Now, here’s where things get interesting. A typical PIR sensor has two, sometimes four, sensing elements, these pyroelectric detectors are responsible for picking up heat changes in the environment. The Fresnel lens acts as a segmented lens, meaning it is made up of multiple small sections, each of which focuses infrared radiation from different directions into the PIR sensor.

What does this mean in practice? The Fresnel lens effectively divides the detection area into multiple zones. An object moving from one zone to another triggers the sensor, helping the camera differentiate between static heat sources and actual motion.

How Do You Choose the Right Fresnel Lens?

Different camera brands design their Fresnel lenses differently, which means each model detects animals in slightly different ways. Some shapes may be better suited for detecting larger animals, while others might be more effective for small or fast-moving creatures. The key principle is that a heat source (an animal) should move from one detection zone to another for the PIR sensor to recognize it as motion.

Here’s a visual comparison of three different Fresnel lens designs:

• The first pattern - lens taken from one Spypoint cameras (can't remember exact model).
• The second design is from the Boldygard MG-series.
• Final - similar to the Spypoint is our own.

If you want to dive deeper into the technical details of PIR sensors and Fresnel lenses, I highly recommend checking out Winterberry Wildlife’s deep tech guides. They provide extremely useful insights into how trail cameras work, including a detailed breakdown of PIR sensors and lens designs.

?? Read more here: Winterberry Wildlife - PIR Sensors & Fresnel Lenses

Battery and Power Management

Well, you guessed it, please give me a solution where I never have to worry about batteries!

Honestly, we’re not far from that reality. With solar power extensions and improved rechargeable battery technology, there are already low-maintenance solutions available. If you're building wildlife cameras with sustainability in mind, consider using rechargeable batteries instead of disposable ones.

For our Animal Detect cameras, we used rechargeable batteries, but we never got around to building a solar-powered extension, which is still something I wish we had done. ?

Solar Power – the good and the bad

While solar panels sound like the perfect solution, they come with trade-offs. One major issue is reflections and glare, solar panels can give off reflections that might alert both animals and people to the camera’s presence. Depending on your use case, this can be a security risk or just make it harder to observe natural animal behavior.

Environmental Challenges and Durability

There isn’t much more to say here, you’re dealing with outdoor environments, and wildlife cameras need to survive in some of the harshest conditions.

• Some animals (looking at you, elephants!) actively try to destroy cameras, so making the power system robust is essential.
• Extreme temperatures can cause battery failure, choosing batteries that can withstand both intense heat and extreme cold is a huge advantage.

Our Solution – Quick-Swap Battery Packs

Instead of overcomplicating things, we designed a battery pack that could be swapped out in seconds. One spin, one pack out, pop another one in, spin and done.

The Brain

We decided to go with an ultra-low-power system to handle the sensors and control everything, a 328p-based PCB that we developed ourselves. While there are many dedicated wildlife camera chips, we wanted more flexibility, so we opted for something we could fully control and customize.

If you’re just getting started and want (close to) unlimited customization, using an ESP32 or Arduino-like chip is a great way to go. These microcontrollers are familiar, easy to program, and well-documented.

But here’s the catch, creating a cellular-connected wildlife camera based purely on one of these chips is hard.

If you need crisp images, 4G connectivity, or even over-the-air updates, you’re going to hit limitations quickly. And if you want AI capabilities, well... good luck running that on a basic microcontroller! (Even if some microcontrollers these days offers some promising features, we may actually start seeing something working quite well soon)

Our Early Attempts – A Pure 328p-Based Camera Board

That doesn’t mean we didn’t try. We actually played around with a 328p-based test setup using two Arducam Mega 3MP SPI Camera Modules, one RGB and one NoIR, with images saved to an SD card.

The upside? Snapping images was basically instant while the cameras were powered on.

The downside? Keeping the cameras powered continuously drained more power than the entire system.

So, we tried using a relay to turn the cameras on and off. This worked, but it introduced some delays. Not the worst, but still noticeable.

Then there was the memory issue. Taking the image was quick, but writing to the SD card and uploading to the cloud? That’s where things started falling apart. I won’t even get into how painful that process was. And while 3MP on an Arduino camera is impressive, it’s not exactly comparable to today’s 12MP+ cameras.

The solution? Raspberry Pi & 4G + 328p board

In the end, we took the easy route, we added a Raspberry Pi Zero W2 to handle image capture, processing, and cloud uploads. To enable real-time connectivity, we also added a 4G module.

The result? It worked, but it wasn’t exactly fast. Our system had a painful 10-second delay from detecting movement to taking and sending an image. That’s fine for feeding stations or situations where animals stay around for a while, but if you’re trying to capture fast-moving wildlife, this is way too slow.

If you’re serious about competing with commercial wildlife cameras, please, keep the trigger time under 0.5 seconds. Don’t repeat our mistake and end up with a 10-second delay, it’s just not practical for real-world use.

What’s Next?

I didn’t get to continue much beyond this stage, but I was introduced to the Rockchip RV1126 chipset, which sounded extremely promising with built-in AI capabilities and ultra-low power modes. At least, that was the sales pitch... I never got around to actually testing it.

Maybe someone else can give it a try? ??

Test Your Camera in the Most Extreme Conditions!

Would you think your hardware might behave differently when exposed to extreme heat or cold? Well, you’d probably be right!

We learned this the hard way. Some of our components, including our 328p board, completely failed when exposed to cold conditions. And when did we discover this? Only after sending it to a tester and a potential buyer. Not exactly the best timing!

Could this have been avoided? Yes, absolutely. Maybe a simple freezer test would have saved us from an embarrassing moment.

Safe to say, testing in a freezer became a standard part of our process after that. ??

What About Heat?

While we never had a way to properly test for extreme heat, we also never encountered issues with overheating, at least here in Denmark. Of course, if you’re designing a camera for use in desert environments or tropical climates, this is something you’ll want to actively test for.

Lesson learned: Don’t assume your hardware will survive just because it works fine in a lab. Test it in realistic conditions, and if you can, push it to the extreme

What’s Missing?

Plug & Play + An Amazing App

If you’re building an online camera, your app needs to be as simple and intuitive as possible, while still allowing users to tweak all the necessary settings. One thing we heard repeatedly from users was that intuitiveness matters, especially when they want to quickly adjust settings.

The dream setup? As close to plug & play as possible. Ideally, the user just inserts the battery, performs a few basic setup steps, and is ready to go. Our goal was to make it as simple as:

1. Open the app
2. Click "Add new device"
3. Scan a QR code
4. Done!

Casing – More Than Just Protection

A good casing should camouflage the camera, protect against moisture, and keep water from reaching any electronics. In other words, something better than what we made! ??

I have no further comments on camera designs as design has never been my strong part.

Antennas for Online Cameras

If you’re designing a connected wildlife camera, antennas can become a big deal, and I’m not joking. They break all the time.

• Birds peck at them.
• Monkeys (yes, really) rip them off or unplug them.
• Even regular wear and tear causes failures.

We experimented with ceramic antennas that we placed inside the casing to protect them. While they technically worked, they just weren’t as good as external antennas. Unfortunately, external antennas are easier to break, so it’s a constant trade-off. And let’s not even talk about the *days I lost researching different types of antennas, their orientation, and omnidirectional vs. directional designs…cries. We ended up with "locking" our antennas which so fat have worked perfectly.

GPS

The more features, the better, right? We allowed users to manually set their camera location in the app using Google Maps when adding a device, but why stop there?

Adding actual GNSS (GPS) tracking makes sense for a lot of reasons:

• Anti-theft tracking: if the camera gets stolen, you can find it.
• Accurate placement tracking: if someone moves the camera, you know.
• User convenience: yes, people forget where they placed their cameras (it happens!).

GPS coordinates embedded in image metadata also make it easier to post-process images, generate heatmaps, and analyze wildlife movement patterns.

Photoresistor – when to switch between Day & Night Mode?

Many cameras come with a photoresistor that decides when to switch from day mode to night mode (when to turn on IR lights). However, some models struggle with this transition, resulting in very dark images at dusk or night-mode images when there’s still plenty of natural light.

If you’re picking your own photoresistor, make sure the resistance levels allow the camera to clearly distinguish between night and day.

We went all-in and gave users 1024 adjustable values, defaulting to 630, which is almost right in the middle. This meant users could fully control when the camera switched to night mode, or even force it to always stay in day or night mode.

Sounds great, right? Well… not really. While it gave maximum flexibility, it caused more problems than it solved. Many users didn’t fully understand how it worked, leading to confusion and frustration. Sometimes, too much adjustability can actually be a bad thing.

Sensors? Yes, Please!

If you can add more sensors, do it. Temperature, CO?, humidity, pressure, each one provides valuable data that can enhance wildlife monitoring. I could probably write a whole separate post about the benefits of these, but just know that adding environmental sensors can help with:

• Understanding animal behavior patterns in relation to weather conditions
• Detecting microclimate changes in wildlife habitats
• Monitoring CO? levels in enclosed spaces for conservation projects

If you’re building a camera, why stop at just images?

Other Means of Detection

PIR sensors are great, but sometimes they’re too general. What if you only want images in very specific situations?

Take feral cat rescue projects, for example. Instead of receiving thousands of unnecessary images, you may only want an alert when the cat is actually captured, with an image attached to verify it’s not a false positive.

We solved this by adding a magnetic sensor to our camera. Just attach a string to a door or trap, and bam, you now have another detection method!

Other alternatives to PIR include:

• Proximity sensors – Detect movement without relying on heat
• Distance sensors – Capture objects at a specific range before triggering

There are plenty of ways to improve how and when your camera activates!

Mesh System – A Wild Idea? Maybe Not.

Apparently, people love the idea of online cameras but absolutely hate paying for multiple SIM cards. Also, cellular coverage is limited in remote areas, so how do we extend GSM connectivity?

Well… I really wasn’t kidding when I said I spent weeks researching this.

A mesh network could extend the range of camera deployments and, in some cases, eliminate the need for satellite communication. One technology that caught my attention was Sub-1 GHz Wi-Fi, also known as “HaLow”.

• Long-range Wi-Fi—potentially 1 KM+
• Excellent penetration (keep it clean, we’re talking about signal strength)
• Low power consumption, potentially lasting years on a coin cell battery

Imagine a chain of cameras relaying data over long distances until reaching a base station with GSM connectivity.

But I see where this is going… let’s get back on track.

AI – We Need More of It

If you’ve been paying attention to the world lately, you already know—we need more AI!

The number of ideas people have..., honestly, it’s fantastic:

• “Can you make an AI that only sends images of specific animals?”
• “Can you use AI to remove empty images?”
• “Can you make an AI that helps me with my Bitcoin investments?”

…Okay, maybe the last one isn’t relevant here, but you get the idea.

That’s actually why we started the Animal Detect Platform, to bring AI to wildlife monitoring and automate some of these processes.

Modularity – Repair or Replace?

One of the things we experimented with in Animal Detect was modularity. Instead of having a single, monolithic board, what if you could upgrade specific parts without replacing the entire camera?

For example:

• Want 5G instead of 4G? Swap the module.
• Broke your IR board? Replace just that part instead of the whole camera.

I love modularity, and seeing more companies adopt repair-friendly designs makes me happy! But of course, there’s a trade-off, do you really want people messing around with the hardware? What happens with warranty claims if users start swapping out parts and breaking things?

It’s something to think about.

Other Random Notes

• Spectral Imaging? – Not sure what I was thinking, but it’s in my notes.
• Thermal Imaging? – Yes, obviously, thermal cameras would be great.

So, what else am I missing? Probably a lot. But to keep this short (wait, I’m already past 4,000 words?!), let’s wrap things up.

If you’ve read this far, congratulations, you’ve made it to the final part of this article! ??

WHY DID WE STOP BUILDING CAMERAS?

Yes, this is probably the most important part of this entire article. If you’re considering going down this road, take time to reflect on everything I write here.

I strongly encourage people to keep pushing technology forward, we need more solutions, more innovation, and more ideas. But I won’t lie, it’s a rough path ahead, and if you choose to walk it, you have my deepest respect.

Building Hardware is Hard. Outdoor Hardware? Even Harder.

There. I said it.

I’ve always been fascinated by hardware development, but my background has mostly been in working with existing hardware, robotics, and software. So diving into building a fully custom outdoor camera was a serious learning curve for me.

Thankfully, our hardware guy, Johnni Nielsen , did an amazing job iterating through several PCB designs throughout our journey. As the CTO, I had to do everything I could to understand whatever was going on with the electronics, especially when things didn’t work.

For those with experience in PCB design, things might go smoother. But just to give one out of many examples of the kind of challenges we faced:

We had a 4G board that kept crashing randomly when booting up the Raspberry Pi and modem.

• Issue? Voltage drop.
• Solution? Let’s throw in larger capacitors everywhere!
• Reality? Nope. The real problem was an outdated IC, and the solution was upgrading the component, not adding bigger capacitors.

We had fully functioning cameras with neat features, but now we faced the next challenge:

• Certifications and compliance
• Two-year warranty (required in most of Europe)
• Servicing and maintenance
• Scaling up manufacturing
• Waterproofing in a country where it rains 100+ days a year (water finds a way into any camera, including ours)

And just to match existing solutions on the market, we would have needed to:

• Drop the Raspberry Pi (too slow, too power-hungry)
• Reduce trigger time (we needed to get under 0.5 seconds)
• Design custom molds
• Set up an entire supply chain

At this point, we were staring at a massive hill still to climb just to make the camera viable at scale.

The Hardest Realization – Was It Even Worth It?

This part hurts. But it was the truth. Do I regret it? Not at all!

Even with all the work we put in, adding AI capabilities, unique features, and optimizations, I had to ask myself a hard question:

Would I personally buy our own camera instead of a well-established brand that people already trust and love?

Sadly… no.

The only thing I was truly proud of was the AI. And it was at that moment we realized something:

?? Why struggle with manufacturing when we could focus on AI and software instead?

We could outsource the hardware manufacturing, the painful part for us, and instead build the software that connected these cameras to our cloud AI.

The Big Questions You Should Ask Yourself

If you’re considering building your own wildlife camera, take a moment to really think about these two key questions:

1. Does your product bring enough additional value to make people switch from established brands?
2. Do you need to reinvent the wheel, or can you build on existing hardware?

The Turning Point for the Animal Detect Camera

We tried hard to push our Animal Detect camera to potential users, but we kept hearing the same thing:

"We don’t need another camera… we’re happy with what we already have."

Ouch. But then they added something interesting:

"What we really need is something to help us manage all the data these cameras produce. Oh, and also our drone footage. And sometimes underwater cameras too."

And that’s how the Animal Detect platform was born.

Instead of fighting an uphill battle with hardware, we pivoted to solving the real problem people were asking for, better software for wildlife camera data management, AI-powered detection, and smarter workflows.

Final Words – If you made it this far, you deserve a Medal ??

It has been a bumpy road, but I encourage anyone who is working on wildlife camera development to keep pushing forward.

Even though we moved away from hardware, I still strongly believe that there is room for new solutions in the market. There are still challenges that need solving.

So if you’re taking this path, I truly hope you succeed!

And finally, thank you to everyone who was part of our Animal Detect Camera journey! ??

[Extra Extra Extra 19-02-2025]

I will continue updating this guide based on ideas and feedback from experts in the field.

Battling Camera Vandalism and Theft

While I briefly mentioned the issue of elephants damaging cameras in a previous sections, vandalism and theft are far more widespread problems. Here, we’re not only dealing with wildlife but also people. WILDLABS has two insightful articles discussing these challenges and potential solutions:

?? How to Stop Thieves When All We Want is to Capture Wildlife Action

?? Solutions to Camera Trap Theft

Additionally, Michael Johnston suggested an innovative approach: separating the camera and storage. In this setup, the camera offloads captured images and videos to a hidden or buried storage device nearby. Even if the camera is damaged or stolen, the data remains safe. This could be achieved through Bluetooth, a hidden WiFi hotspot, or another wireless connection, where the camera acts as a transmitter and the storage device as a receiver.

To tackle issues with animals interfering with cameras, such as muskoxen and polar bears in Greenland, Lars Holst Hansen proposed integrating a LoRaWAN tilt sensor. This sensor would detect sudden changes in yaw, pitch, or roll and send an alert when the camera is displaced. This idea could also serve as a security feature to notify researchers if people are tampering with the camera.

Synchronized Date and Time

Most cameras require manual setup for date, time, and time zone selection. Many offline cameras default to 1970-01-01 00:00:00, following the Unix Epoch time standard. While GPS/GNSS signals can provide accurate timestamps, there are other solutions for online cameras.

With the Animal Detect Camera, we used GPS coordinates from the camera's location to determine the correct time zone and then pinged a time server to synchronize the clock, ensuring precise timestamps down to the second.

Looking back, one improvement I would make is embedding both the datetime and GPS coordinates directly into the image metadata. This would be highly beneficial for post-processing tasks such as creating heat or density maps, where accurate time and location data are crucial.

Price and Affordability

While I haven’t previously discussed price and affordability, they are critical factors when designing trail cameras. Who is your target audience, and what are their financial constraints?

For instance, if you’re building cameras for conservation projects in Africa or Thailand, cost becomes a key consideration. Cameras frequently break, and funding in these regions is often limited, relying heavily on NGO grants and conservation budgets. Keeping costs low while maintaining functionality is essential in these cases.

Timelapse vs. PIR-Sensor Cameras

Heinrich S. suggested using a timelapse mode, where the camera takes an image every minute, rather than relying on a passive infrared (PIR) sensor to trigger shots. This approach has some interesting benefits but also comes with challenges, particularly in data management, where AI could play a crucial role in filtering and processing images efficiently.

Personally, I love the idea of a timelapse camera. I often wonder what and how many animals a standard trail camera setup fails to capture. A timelapse system, combined with AI, could provide insights into smaller or more elusive wildlife:

? Amphibians, small birds, and smaller mammals

? Animals that are too far away for PIR detection but still visible in images

? Wildlife active during known peak times but not captured by traditional setups

In some African conservation areas, researchers report periods where they know wildlife is present during the day, yet their cameras don’t capture any images. This could be due to PIR sensor limitations, or otehr issues like overheated electronics and battery failures.

A timelapse + AI approach could be an exciting way to compare detection rates and fill in the gaps left by standard trail camera methods.