The intelligent visual recognition edge-computing software technology achieves the integration and optimization of 7 functions of AI-ADAS detection and recognition (traffic lights, lane lines, driving areas, four-wheeled vehicles, two-wheeled vehicles, pedestrians, signal status, and semantics) on the embedded platform. The real-time computation speed reaches 30 FPS, and the performance improved by 2.63 times. Based on four commercial chips (Qualcomm, Renesas, iCatch, Jetson), the cost-effective distributed sensing platform can be constructed with the cost reduced to 13％ of the original price.
The integration with innovative deep learning architecture can further improve the performance by 1.65 times. This technique has been introduced in commercial applications such as real-time lane shift (LDWS) and front collision avoidance warning systems (FCWS).