Engineers Audi Electronics Venture created a car camera with artificial intelligence to create high-precision three-dimensional models of the surrounding space. Audi specialists are sure that the complex has a great potential for interpretation of road situations.
It has an overview of 120 degrees and makes 15 shots per second with a resolution of 1.3 megapixels.
The received images are processed by a neural network using semantic segmentation, during which each pixel is analyzed and distributed over 13 classes of objects. This allows the system to identify and distinguish other cars, buildings, road markings, people and signs.
In addition, the system solves the problem of determining the distance. To do this, an image is constructed using isometric lines, which are virtual boundaries that determine the constant distance between objects.
To train the neural network, a lot of video recordings of traffic situations using a stereo camera were used. During the tests the system learned to learn the rules of constructing a three-dimensional picture on the basis of images from a mono camera.
