針對安全攸關之嵌入式即時系統的異質性資訊融合

Abstract

Nowadays, the bulk of these road collisions is caused by human unawareness or distraction. Since the most important thing is your safety and the safety of others, ADAS is developed to support enhanced vehicle system for safety and better driving. AEBS as an important part of the ADAS has become a hot research topic. Computer vision, together with Radar and Lidar, is at the forefront of technologies that enable the evolution of AEBS. Since the cost of long range radar and lidar is very high, we want to use camera-based system to construct AEBS. Instead of using a single monocular camera, we propose a heterogeneous camera-based system to use sensor fusion to combine the strengths of all the difference FoV cameras. Also,We use a heuristic false positive removal method to decrease the false positive rate that caused by the sensor fusion method. We optimize the sensor fusion method Because of the the limitation of computing resource on embedded system. As a result, the recall of YOLO can be increased up to 10% through our heterogeneous camera-based system.