3D thermal ranging for pedestrian detection and identification versus RGB in compromised lighting environments. Filmed on the streets of Las Vegas in Jan 2022.
[Transcript]
Hello this is chuck gershman, CEO at OWl AI
A study released last month from the insurance institute for highway safety showed that today’s production-grade pedestrian automatic emergency braking systems reduced crashes by up to 32 percent during daylight hours however the same study showed absolutely no difference in pedestrian crash incidents at night effectively a total fail at night shot at night on the streets of las vegas during ces 2022 in january
This video demonstrates the unique value proposition of incorporating 3d thermal ranging capabilities to pedestrian detection systems for dramatically improving safety at night this frame synchronous synchronized video shows a video stream from a visual camera side by side with a video stream from al’s new 3d thermal ranging solution I was applied the same object classification algorithms to both videos using blue boxes to identify vehicles and red boxes to identify pedestrians al’s 3d ranging aiml algorithms are also applied to the thermal stream on the right we present the range in the form of a color gradient to represent depth from the camera the color wheel shown on the bottom going from red to purple represents near field to far field this color wheel is mapped automatically by the software onto the video to present range to the objects of interest
Now we’re going to take a look so let’s run the video and i’ll do that slowly and as we go we’ll move forward a few frames and get the cars moving and immediately we see a couple of pedestrians on the right hand side in the thermal image and they’re identified by the red boxes we do not see pedestrians on the left in the visual as they are delayed in it and we have to get much closer for that to occur now as we move forward a little bit further a few more frames we now have four pedestrians identified uh in the thermal and only one and actually here we have four and none now we have four and a single pedestrian in the visual and we have to come pretty far pretty close um in order to get a four and four match so as you can see the thermal has a higher fidelity in terms of its ability to identify pedestrians and it identifies the pedestrians much sooner and this phenomena just continually repeats itself here we see it again with another pedestrian further down and here yet again same phenomena now i want to jump here to and stop you’ll see that there is the front tire um in the front of a car coming out of the drive on the and you can see that clearly on the right hand side of thermal if you look closely on the visual you’ll see the car in the thermal we have already classified it as a vehicle giving it a purple box but not no such case in the visual and as we move forward there the visual just picks up the car now i’m sorry so we will stop right there now notice that we’ve gotten the car now both in the uh thermal and the visual but now the thermal has picked up a pedestrian far far away that’s about 60 meters but no such pedestrian is seen in the visual so let’s see how far we’re going to have to drive the car forward before we get a hit on that pedestrian number frames we’ve come a half we’ve come almost a half a block here still nothing still nothing still nothing and just as we get to the corner we get our first bounding box on the pedestrian individual side we’ve had a consistent bounding box uh in the thermal the whole time so let’s move up to two and we’ll see that it’s intermittent but as we drive all the way up and stop we can see okay now that we can see the image cl a little more clearly we can see that the pedestrian is standing next to a sign and next to a bus stop um this clutter has created a very very ambiguous situation for the for the visual camera in addition we’re getting significant headlight glare off of the bus stop all of these compound the problem including the fact that we’re looking at this at night to make this uh individual pedestrian very very difficult to see um he’s one step from the curb the cars are moving 45 miles per hour at this distance we would not be able to avoid or stop for him safely if he were to step into the curb and it’s just now that we’ve been able to identify him so now you can clearly see what the value proposition of adding a 3d thermal ranging to a pedestrian automatic braking system would bring from a safety perspective we’ll be bringing you more videos in the future on this newsletter thank you for listening you
