The core differences between detection abilities of the two methods
We get asked this question a lot: why would we need a stereo vision system (two cameras) if we can use a mono vision system (one camera only) to detect cars and pedestrians on the road?
The answer is simple. A mono vision system enables detection of classified objects ONLY, such as cars and pedestrians, while a stereo vision system enables detection of, and measures the exact distance to, ALL objects ANYWHERE in the field of view, regardless of their shape, size, and color (even if it’s a UFO that just landed in front of you).
A typical stereo vision system (left) and a mono vision system (right) behind the front windshield
Mono vision systems rely on object classification, meaning that they must be “trained” to detect what certain objects look like. Those objects are categorized as “classes”. Classes may include pedestrians, cyclists, passenger cars, trucks, etc. each in its own separate class. For each captured image, an algorithm tries to match one or more of the “classes” in its library. If the system finds a high certainty match, it can estimate the distance to the known object based on the previous knowledge it has of how those objects are supposed to appear for each distance, and can alert the driver about upcoming obstacles. If it doesn’t find any match, the mono vision system will proceed as if there were no objects at all and will not provide any alerts.
This is the main drawback in mono vision system technology. Sure, it does wonders and looks like magic thanks to AI and deep learning advancements, but the moment it runs into an unclassified object, it’s practically blind and acts as if there is nothing in front of the vehicle. This was demonstrated numerous times in reported accidents of semi-autonomous vehicles.
For this reason, mono vision systems cannot be used as the sole technology sensor in future autonomous vehicles. These systems must be combined (or “fused” in technological terms) with other technologies such as LiDARs and radars. (More on that in a future post.)
Stereo vision systems overcome this serious shortcoming by using inherent 3D capabilities that enable them to see all objects around them, just like our eyes do, even if the system wasn’t trained to “know” what the object is.
This is the uniqueness of the QuadSight® vision system. The QuadSight system consists of 2 pairs of stereoscopic vision channels: visible light, and thermal. When the object detection capabilities of both stereo channels are seamlessly fused, the result is unparalleled object detection in harsh weather and lighting conditions. QuadSight can produce a full 3D map of the surroundings in day and night, sun and rain, and even very foggy or snowy days. It can detect ALL obstacles or people that might interfere with the vehicle’s driving path.
QuadSight is truly a milestone in the autonomous vehicle revolution!