China develops intelligent driver assistance system

According to local media, the Second Department of Second Research Institute, CASIC (China Aerospace Science & Industry Corp) has successfully developed a deep-learning-based intelligent driver assistance system that can perceive the real-time surroundings by virtue of the tiny embedded chips. The accuracy in object identification has reportedly reached the world-advanced level.

The public data show that the intelligent driving system's highest accuracy in object identification has so far reached 90.55%, while it still takes 4 seconds to process an image. The algorithm of the driver assistance system developed by CASIC can reach an accuracy rate up to 90.05%. What's more, it only requires 0.03 seconds to process an image.

The R&D team of the Second Department has made a series of breakthroughs in many core technologies, including multi-object detection and identification, driving area segmentation and traffic lane detection, etc. Besides, the team has built technical barriers in such areas as deep neural networks (DNN) compression, DNN compiler toolchain and intelligent accelerator, etc. Guo Rui, director of the R&D team, explained that the scene semantic segmentation is to let computer understand what the image it sees stands for. Using DNN, the machine can extract the high-level semantic features by self-learning which exempts the object identification from being interfered by backlighting, shadow and image defect, etc.

Guo Rui added that they have shifted the working focus to the engineering and production of the intelligent driver assistance system. Besides, the team has joined hands with some automakers to test the product's function and conduct trial-manufacturing. It is expected that the intelligent driver assistance system will be mass produced on a small scale by the end of this year.

Additionally, the R&D team is developing a new-type intelligent sensor that integrates visible light, infrared and millimeter-wave radar, which is available even in some special scenarios, such as complex light conditions and the night march with lighting forbidden.