A few days ago, the second division of China Aerospace Science and Industry No.2 Institute made new breakthroughs in the area of artificial intelligence products. It successfully developed a prototype of a “deep learning-based intelligent assisted driving system” function prototype. This product can be transplanted into embedded chips with a size of a few centimeters. It can realize real-time accurate and intelligent perception of the environment and reach world advanced level in accuracy of target recognition. On the smart driving public data set KITTI, the general target recognition accuracy rate is up to 90.55%, and processing a picture takes 4 seconds, but The intelligent driver's algorithm can achieve an accuracy of 90.05%. It takes only 0.03 seconds to process a picture. The system consumes less than 5 watts, which can effectively reduce rear-end crashes, head-on collisions, and road-related accidents.
At present, the team has broken through a series of core technologies, including multi-target detection and recognition, segmentation of driving areas, detection of lane lines, etc., and has formed technologies in such areas as deep neural network compression, deep neural network compiling tool chains, and intelligent acceleration engines. Advantage.