Radar-Camera Fusion Method for All-Weather Traffic Perception in Complex Scenes

This paper proposes a decision-level fusion approach to effectively integrate multi-source heterogeneous sensor data to meet the challenge of all-weather traffic perception. First, considering the effects of illumination and climate factors, we encode the two types of information to guide the adaptive fusion process of heterogeneous sensor data in different scenarios. Second, we transform the radar and camera fusion task into classification and regression tasks. The transformation facilitates implementing the fusion perception using existing classifiers and regressors with coded information into the model. Finally, we conduct experiments on a homemade data set containing four kinds of scenes, e.g., sunny, foggy, dusk, and night. The results demonstrate that the proposed method can greatly improve the accuracy of vehicle detection, compared to traditional single-source sensing methods. It shows the potential significance of intelligent traffic parameter estimation and traffic safety monitor.