Automotive

ShadowCam looks for movement in the shadows

ShadowCam looks for movement in the shadows
By detecting movement in the shadows, MIT's ShadowCam could help autonomous vehicles avoid bumps with vehicles coming around corners
By detecting movement in the shadows, MIT's ShadowCam could help autonomous vehicles avoid bumps with vehicles coming around corners
View 1 Image
By detecting movement in the shadows, MIT's ShadowCam could help autonomous vehicles avoid bumps with vehicles coming around corners
1/1
By detecting movement in the shadows, MIT's ShadowCam could help autonomous vehicles avoid bumps with vehicles coming around corners

Autonomous vehicles see the world around them using lasers and cameras, with captured data run through powerful computer systems to help them get around without incident. But there's always room for improvement. MIT engineers are developing a system that looks at changes in shadows on the ground to avoid collisions with objects appearing from around corners or behind buildings.

"Our method can give the robot an early warning that somebody is coming around the corner, so the vehicle can slow down, adapt its path, and prepare in advance to avoid a collision," said co-author of the study, Daniela Rus.

The early warning anti-collision system builds on an earlier project dubbed ShadowCam, where a camera is pointed at the floor and changes in light intensity are detected. Some of these changes may be too subtle for the human eye to detect, but could be the first sign of an object (such as a pedestrian or vehicle) appearing from around a corner.

The original made use of augmented reality tags positioned throughout an area to help it look for movement in shadows, but this isn't very practical for real-world applications. The updated ShadowCam technology combines the overlaying of multiple images to detect changes in light intensity and real-time analysis of camera motion that can use feature points in an area instead of relying on AR labels.

The system looks for subtle differences in the target area, gives any suspected shadows a color boost to reduce the signal-to-noise ratio and determines if the object causing a shadow is moving or not. If it is, ShadowCam instructs the autonomous vehicle to slow down or stop.

The MIT team has only tested the system indoors so far, at slow speeds and with consistent lighting. In one such test, an autonomous wheelchair moved towards hallway corners while humans turned the corner into its path. The system was tested with AR tags installed in the hallway and without, and was found to achieve the same level of accuracy at classifying moving or stationary objects in both cases.

The ShadowCam was also installed in an autonomous car and tested in a parking garage set up to mimic driving at night. Detection times where compared with a vehicle using LiDAR and the ShadowCam was found to be 0.72 seconds faster at registering an approaching car. That may not be much in slow-moving traffic, but could be the difference between accident and safe journey in fast-movers. Classification accuracy also improved compared to the wheelchair tests thanks to the researchers tuning the system to the garage's specific lighting conditions.

A paper detailing the current system is being presented at the International Conference on Intelligent Robots and Systems next week in Macau, China. Work on the ShadowCam project continues, with the researchers announcing that the next steps include getting the system to work in different indoor and outdoor lighting conditions, as well as improving on detection speeds.

Source: MIT News

No comments
0 comments
There are no comments. Be the first!