Frontal object perception for Intelligent Vehicles based on radar and camera fusion

This paper addresses the issue of frontal object perception in real-world traffic scenarios. Accurate and real-time frontal object perception plays a key role in Advanced Driver Assistance Systems (ADAS) and Intelligent Vehicles (IV). However, perceiving complex traffic environments, which contain multiple classes of on-road objects with various visual appearances from different viewpoints and partial observations, is still a challenging task. In this paper, a perception system fusing a millimeter-wave (MMW) radar and a monocular camera is proposed. Firstly, the detections of MMW radar are converted to regions of interest (ROIs) on the image. Then, these ROIs are classified by four classifiers using Deformable Part Model (DPM). Finally, a mixer module is used to combine all the classification results and infer the final result for each ROI. The computation intensity of the DPM algorithm can be efficiently reduced through this mechanism. Meanwhile, high detection precision is achievable. Experiment results show that the proposed frontal object perception system can detect and classify on-road objects in complex urban traffic scenarios with 98.4% detection rate at nearly real-time performance (29Hz).