A Lightweight Motional Object Behavior Prediction System Harnessing Deep Learning Technology for Embedded ADAS Applications

This paper proposes a lightweight moving object prediction system to detect and recognize pedestrian crossings, vehicles cutting-in, ahead applying emergency brakes based on 3D Convolution network for behavior prediction. The proposed design significantly improves the performance of conventional convolution (C3D) adapted predict behaviors employing recognition capable performing localization, which is pivotal in detecting numerous objects’ behaviors, combining verifying detected objects with results YOLO v3 detection model that C3D model. Since CNN requiring far lesser parameters, it can be efficiently realized an embedded real-time applications. achieves 10 frames per second (FPS) NVIDIA Jetson AGX Xavier yields over 92.8% accuracy recognizing crossing, 94.3% vehicle cutting-in behavior, 95% brakes.