American engineers have created a system that allows you to detect oncoming objects due to the angle of their shadows or other changes in lighting. Developers have tested this system on an unmanned vehicle, another car will be able to brake before leaving the wall and crossing the road. The article will be presented at the IROS 2019 conference and MIT News spoke briefly.
In the field of unmanned vehicles, the concept applies to the use of a number of sensors operating on different principles: lidar, radars and cameras. This allows you to use the detection capabilities of each type and combine the advantages for reliable monitoring of objects as well as replication of systems.
However, all these sensors are only used to fix objects directly in the line of sight, while people use indirect signs that indicate that the object is out of sight. For example, a driver can predict that a headlight can leave behind a building after seeing a shadow during daylight hours or vice versa, and can also hear the sound of the engine and tires.
Researchers at the Massachusetts Institute of Technology and the Toyota Research Institute, led by Daniela Rus, have proposed direct observations of objects based on assumptions based on such indirect properties. So far, an article describing the system and experiments has not been published, but some conclusions can be drawn from the article in an earlier version of the algorithm.
The algorithm takes video from the camera, basically creates a three-dimensional model of the media in the frame and reflects the frames. The system then independently selects the region of interest, for example the corner of the wall. Then, as the wall approaches, the algorithm monitors and analyzes this specific area. For analysis, saturation increases in images, which makes it possible to increase the difference between the shade and the illuminated area of the floor. The algorithm then determines whether the shadow will move and can then decide to stop.
In a recent study, engineers tested the system on an unmanned vehicle, not on a robot. In the published video, you can see that the car foresaw another car approaching to change the car's lighting. Developers said the algorithms detected a moving obstacle in lighting.
There are other approaches to tracking obstacles around the corner. For example, last year Honda showed the work of a smart intersection prototype in the US. It is equipped with cameras that recognize cars approaching the junction, as well as antennas that transmit data about them to other nearby vehicles. This allows drivers to know in advance that another car or pedestrian will appear in the corner.