On-road vehicle detection fusing radar and vision

This paper presents a cross-verification approach to fuse radar and vision data for vehicle detection. Firstly, a realtime vision approach using specific shadow segmentation is used to detect vehicles in whole image independently. The fusion approach contains two steps: matching and validation. The targets respectively from radar and vision verify each other in matching process. Then the unmatched radar targets are validated by vision data once again. Experiment results with test dataset from real traffic scenes on freeway and urban roads are presented to illustrate the performance of this approach.