Scientists from the American Cornell University announced work to create a new way by which unmanned vehicles can receive information about the environment.
Thus, it is noted that the system uses stereo cameras and neural network algorithms. The developers claim that it will be much cheaper than the currently used lidars. Lidars scan the area around the car, determining the distance to possible obstacles using laser beams and creating a so-called “point cloud”. According to one of the developers of the new technology, Kilian Weinberger, this is an effective, but expensive method: one lidar costs about 10 thousand dollars.
Instead, Kilian Weinberger and his colleagues propose to use the so-called “stereo vision” based on two video cameras spaced about a meter apart. Computer algorithms based on neural networks, comparing images from these cameras in real time, will be able to calculate not only objects in space, but also the distance to them. It is planned to achieve this by analyzing the “displacement” of visible objects relative to the central axis of “vision”, as it happens with human eyes.
“For example, we can determine where a pedestrian is, surround him with a kind of electronic frame and tell the algorithm that there is a person inside the frame, so this place must be bypassed,” the scientist noted. When exactly the conceptual idea will find its real embodiment, there is no information yet: scientists are working on algorithms, as well as machine learning technologies.