You Only Look Once - Algorithm

Imagine that you go blind for a day. I know that's pretty "dark" but bare with me for a minute. It's going to be hard for you to cope up with your regular activities and surroundings, right? Similarly, machines that are meant to be driven by humans need to see if they need to function on their own. Well… I am talking about autonomous vehicles that use computer vision as their eyes.

Here is a Skill Pill for computer vision specifically on YOLO Algorithm. It's not the ‘You Only Live Once’ phrase, but it's ‘You Only Look Once'. Lets get to know what this algorithm does in the autonomous vehicle industry.

What is YOLO?

YOLO is an algorithm which uses Neural Networks for real time object detection and recognition. If you're wondering what Neural networks are, then let me tell you what it is. A Neural Network is a part of Machine Learning and Deep Learning where algorithms try to interpret data in a way similar to the human brain.

In an autonomous vehicle, a camera is placed in a specific position/location inside the vehicle for acquiring the live feeds. The camera acts as eyes for the vehicle but can't understand on it's own because it's a machine. It needs a support for this to understand this on their own.

When growing up, our parents told us that the object we held in our hand was a ball. If they told me it was a cat, I would have ended up referring to it as a cat rather than the toy ball. The same applies to autonomous vehicles. The live feed from the camera is given to pre-trained datasets for understanding the objects. This "Pre-Trained" dataset is basically the parent and the computer is the baby. This dataset helps YOLO algorithm to recognize the object in a few milliseconds (depending on the processing speed of controller/processor).

Working of YOLO Algorithm - for Object Analysis and Detection

Now let's take a look at how this works. Consider this image of traffic as shown below

1. The algorithm splits the image into several segments and each segment is analyzed.

2. All the segments add up to generate a sort of map. Individual segments are responsible for scanning parts of the image and each segment is given weights/labels for recognizing the object and this is stored in a matrix

3. Mathematical functions called activation functions are used to get an output from the matrix. The algorithm then predicts 5 parameters namely, the x and y co-ordinates, width, height, and confidence components.

4. The number of bounding boxes on the screen now reduces to a smaller number based on a minimum confidence score threshold assigned to each. Then you see the output as shown below where bounding boxes show up on each detected object.

YOLO isn't the only algorithm in town. In fact, there are others with better accuracies than YOLO. But, YOLO has an edge over them by being faster with improved versions (like YOLO V3 and YOLO V5).

The human eye detects and processes images in 13 milliseconds [1] but Computer Vision techniques do it even faster.?This makes the autonomous vehicle practically faster than a human.

The combination of LIDAR sensors and YOLO allow cars to visually navigate through roads, and dense traffic. This helps in reducing and avoiding car accidents happening every year all over the world. [2]

Here is what we did with our in house team of technical members who use YOLO for autonomous vehicle. We took an image of traffic and used computer vision to show you how it exactly looks like. (Fig 5 and 6).

Why YOLO?

To re-emphasize on why YOLO is popular, it is fast and can use multiple pretrained datasets for object recognition.

• Researchers have claimed YOLO to be fast in object detection/recognition because of its ability to process 155 frames per second.
• Inside every grid cell, YOLO would predict unique elements like locations, sizes, and confidence scores.
• The confidence score indicates the accuracy and prediction percentage of the object.?

Well, that's it for this Skill Pill on YOLO. There are many more algorithms in the field of computer vision that deals with AV's. Skill Lync has got you covered if you want to know more about getting the perfect industry-relevant courses to get you trained on these topics. https://bit.ly/3TW60st

Reference Links:?

1. https://news.mit.edu/2014/in-the-blink-of-an-eye-0116
2. https://www.cdc.gov/injury/features/global-road-safety/index.html
3. https://towardsdatascience.com/object-detection-with-yolo-bringing-vision-to-self-driving-cars-980295226830
4. https://www.geeksforgeeks.org/yolo-you-only-look-once-real-time-object-detection/
5. https://appsilon.com/object-detection-yolo-algorithm/