Camera Vs RADAR for AEB

One of the key question to be asked while designing any FCWS system is whether to use camera or RADAR for detecting obstacles.

The first key point to understand while designing FCWS is it is not an eyes off , brain off , mind off type of a system and that means driver is in full command of the vehicle all the time .

Forward collisions typically happen when either driver is not in a position to take a correct decision ( for example he is under influence of alcohol) or driver is not able to take correct decision ( for example - pitch dark night) or some vehicle appears sideways on cross junction where driver cannot see (blind spots)

If we wish to use only?#camera ?as a mechanism to detect obstacles, it poses lot of challenges algorithmically as (monocular) camera does not provide any direct mechanism for depth measurement, and it only provides an indirect way of velocity determination. What it does is estimation instead of measurement. Typical approach involves using deep learning model or geometrical model for estimation of depth both have issues associated with accuracy, which gets improved only with time and this means the system is initially under performing and point of singularity is hard to determine.

Camera based systems also suffer with same sensory limitations as human eyes like not seeing anything in pitch dark nights , in fog , in heavy rain etc.

Last but not the least just on sensory grounds camera is far inferior to human eye , in order to make it match with human eye , we must have the camera with following specifications .

1. 576 Mega Pixel Resolution
2. Ability to distinguish 1 million color shades
3. 220 frames per second or more
4. Stereo Camera for depth perception
5. Mechanism to rotate the lens
6. Focal Length of 2.5 cm
7. Horizontal FOV: Area of Focus of 60 Deg , Peripheral vision of 30 Deg , full range of binocular vision of 120°
8. Ability to see in lux levels as low as 0.1 lux ( most lux meters can not even measure this)
9. A processor which can process such an image with 576 megapixel in few milliseconds

That said Camera based systems are perfect for object classification and are also low cost , they provide mechanism to record video of the road in front as an evidence in case of an accident .

RADAR on the other hand has an excellent performance in all weather conditions has a longer range than camera, is not affected by direct visual obstruction , basically it can see what human eyes can not see under certain conditions and hence we call it as extra sensor - the sixth sensor !

Further it provides direct way of measuring velocity and distance and is purely based on physics and mathematics. Unfortunately, RADAR based systems have a disadvantage that they cannot classify the objects (including imaging RADARs) to such a precision as camera, neither can they record the scenario and are typically high cost. Cornering performance of RADAR needs development special hardware like beam forming and also intelligent algorithmics which makes these systems complex to design. RADAR based systems also pose several challenges in terms of handling stationary obstacles like boom barrier etc ( it is more relevant for system like AEB )

However, considering the pros and cons for FCWS, RADAR based systems seems to be the better choice with an optional integration of dash camera for event based capture of road video

On the other hand any system which does an active intervention like Automatic Emergency Braking, neither camera nor RADAR based systems are useful , what we need a fusion RADAR to detect point cloud , distance , velocity & camera to classify objects and then a fusion to decide the final control action.