Light Co. announces its automotive 3D depth Clarity platform:
"Lidars do a great job, but they don’t do the whole job. Their range is often limited to ~250 meters. Class 8 trucks need at least 400+ meters to come to a complete stop, safely. Lidar as well as monocular camera-based systems can get confused as to whether they’re seeing a person painted on the side of a truck or an actual person.
Clarity is a camera-based perception platform that’s able to see any 3D structures in the road from 10 centimeters to 1000 meters away — three times the distance of the best-in-class lidar with 20 times the detail."
“There is nothing else like the Clarity platform with its combination of depth range, accuracy, and density per second. It enables a new generation of vehicles that can be made safer, without having to compromise on cost, quality, or reliability,” said Prashant Velagaleti, Chief Product Officer of Light. “Rather than only minimizing the severity of a collision, having high fidelity depth allows any vehicle powered by Clarity to make decisions that can avoid accidents, keeping occupants safe as well as comfortable.”
their solution seems to be a wide-base stereo camera system with some matching algorithm behind.
ReplyDeletehttps://light.co/technology
thus their full operation is limited to the shared view volume their camera set does have. hard to say if the system will work at least to some basic quality in night conditions.
in contrast to that radar and lidar systems are often built and used with wide angles and thus can serve side impact protection as well as providing also good helps at road corners.
The depth accuracy is quite poor and decays quadratically as is typical of stereo systems. And such a wide baseline multi-camera system would be a lot more bulky and potentially more expensive than lidar. And it won't work well for flat, featureless surfaces.
ReplyDeleteHow could the Clarity system could be more expensive? It looks like it is "only" 3 visible cameras plus some electronics and software compared with Lidar that requires a powerful laser source and/or a SWIR camera and is much more complex mechanically...
DeleteOn Light website, they announce to target a 250$ selling price. There is no way an efficient 3D Lidar could reach this price in the couple of years to come.
Lidar may have advantages, but I would not choose the cost argument!
low cost lidar systems are out there. cost reduction was a major topic for e.g. the automotive industry in the past years and has not stopped at present.
Deletein contrast the idea of having a stereo camera set on the roof top is already more than a decade old. a German(?) freelancing inventor demonstrated it once with a roof carrier setup on a passenger car - and also indicated that a more pleasant series solution can be the integration into the front ends of the pair of rail mount bars on the roof top.
meanwhile the industry went for centered, behind the mirror devices that also do integrate the processing unit into a typical ECU housing. they felt more comfortable and price-performance-optimized at that time in having a single unit than a 3-part design interconnected with high speed data cables going to the exterior with all the heavy weather and the chances to obstruction. (at present the world has learnt that even a behind-windshild positioning does need some explicit cleaning so the desired functionality is kept up. the fact that even the indoor temperatures will any now and then need a bit of extra heating device for the view spot on the front screen was clear from nearly the beginning.)
current lidars do buy up to a 360° view at a moderate effort - and their integration into the vehicles is getting less of a mess due ongoing size reductions. the micro-fication attempts are often using alternate, cheaper approaches than the concepts (such as high quality surround scanners) known for a while as used for e.g. geo system. other concpets, other strenghts and weaknesses, different quality of service - its just to make the solution fit the to target application case - some decision is a compromise, some other is a set back in some parameters, some other is a great overall benefit including price, robustness, mechanic metrics and some of the scan metrics.
the competition of the systems has not ended - and often there is the call to have complementary systems in order to grow the vehicles safety level so that we can trust (as we could see in recent years prominent and not so prominent failure cases) into comfort, safety, semi- and full-automated systems. sensors based on radar, lidar, camera and more are able to serve this for the mass market. multi "spectral" combinations are the solution even if the individual selection and arrangement is subject to vendor/model specific selection based upon a quite high multitude of criteria. it might be the case that for e.g. trucks and busses the "light.co" solution is more adequate or indicated then for any other type of vehicle - but then its a niche with lower unit counts at higher runtime and reliability demands.
That's correct. Error grows quadratically with distance and computing resources are cubic with camera resolution. The best pseudolidar solutions today use a lidar to correct for those errors. http://www.cvlibs.net/datasets/kitti/eval_object.php
ReplyDeleteDoes anyone happen to know what happened to this company? All traces (website, youtube videos) seem to have vanished. Their depth sensing results were impressive.
ReplyDeleteDoes anyone happen to know what happened to this company? All traces (website, youtube videos) seem to have vanished. Their depth sensing results were impressive.
ReplyDeleteThey were forced to close their doors a few years ago when Softbank declined to fund them further.
ReplyDelete