Improving ADAS at Night

Photon-level precision for tomorrow’s automotive safety systems

Many modern ADAS systems rely on their cameras to make accurate decisions. Current sensors available today are fairly capable at night, but they still leave much to be desired. Performing well in 90%, 95%, or even 99% of situations simply doesn’t cut it when the lives of drivers, bikers, and pedestrians are on the line.

To show what Ubicept’s photon-level precision can offer, we compared our solution to a higher-end automotive camera. For context, here’s a clip from the automotive camera with typical street lighting:

As you can see, it actually performs quite well. While this situation is far from ideal, the lighting here is still fairly generous and the car is sitting still. Let’s see what happens when we push the limits by driving down a dark, shaded road with only our daytime running lights:

Even with the gain on the automotive camera turned up, it’s clear that there simply isn’t enough light for the video to be useful. In contrast, the Ubicept solution is able to capture all the necessary detail for an ADAS system to perform navigation, collision avoidance, and so on.

Granted, this is a somewhat contrived experiment. Most drivers, after all, would have their high beams on in this situation. Still, the Ubicept solution offers many advantages in common night driving scenarios. Take, for example, this side-facing scenario on a well-lit city street:

At first glance, they both look quite good! But there are some key differences revealed in the freeze frames such as this one:

The image from automotive camera shows high levels of motion blur. In contrast, the Ubicept solution is able to provide a perfectly sharp image thanks to its photon-level precision and sophisticated processing.

But what is the relevance of this, you ask? Is it necessary for a collision avoidance system to read the stenciled number “87” on the lamppost? Perhaps not, but having zero motion blur can be hugely advantageous for downstream perception tasks such reconstructing 3D information about the environment. Take a look, for example, at the result of an off-the-shelf photogrammetry algorithm:

Clearly, the Ubicept solution is able to provide much more useful information resulting in a far more detailed model of the scene! To see more, check out our demos here. And if you’d like to learn more about how Ubicept can improve your ADAS system, please feel free to contact us any time!

Previous
Previous

Perception with Extreme Dynamic Range

Next
Next

Better Machine Vision in Low Light