Car trajectory prediction in image processing and control manners

Abstract

This paper studies car trajectory predictions desired in several active safety systems, such as automatic emergency braking (AEB) systems and lane keeping systems (LKS). In the former, the car trajectory is estimated such that objects detected within this trajectory in front of current host vehicle are taken into consideration of collision avoidance. In LKS, the trajectory predictor is utilized to evaluate a lateral displacement so as to keep the host car running within the selected lane by reducing this displacement. To accomplish the prediction task, a trajectory estimation strategy is newly proposed in a fusion manner. First, a road model - capturing road geometry characteristics and combined with a vehicle dynamics - is referred; it includes two parameters related to road curvature, which are both estimated in a nonlinear manner. An image processing mechanism is alternatively adopted as a compensating curvature estimator if the proposed observer was in transient response. After the evaluation of road curvature, the vehicle trajectory in future few seconds are estimated in a series of path positions for the car mass center. Experimental results show the proposed predictor guarantees at least 95% of estimation preciseness compared to a real path. Target sensing in AEB worked rapidly in real-world tests with the aid of the path estimator.