Performance Evaluation of Visual Object Detection for Moving Vehicle

With the rapid development of intelligent vehicle technology, the evaluation of the perception algorithms is crucial for safe driving of intelligent vehicles. This paper proposes a novel evaluation method for visual object detection of intelligent vehicles. The traditional evaluation methods are based on each frame of the video and treat all the objects equally. Distinguished from that, the proposed method applies the length of driving trajectory to evaluate the detection algorithm. Besides, the proposed method brings about the road region constraint, weight of safety and weight of early detection. The road region constraint ensures that only the objects relevant to intelligent vehicles are taken into account. The weight of safety for each object is determined based on its degree of relevance of the safe driving. The weight of early detection is set to evaluate whether the objects are detected in time or not. In experiments, the proposed evaluation method can reflect the relatively true detection ability of intelligent vehicles compared to the traditional methods.