Highway Traffic Flow Estimation for Surveillance Scenes Damaged by Rain

In this paper, we propose a traffic flow estimation system for intelligent highway surveillance applications under rainy conditions. Major contributions of the proposed system include flexible feature extraction, robust estimation with adaptive clustering, and effective graph-based traffic flow mapping model. To detect rain-drop tampered scenes, features are extracted via salient region detection and block segmentation. For traffic flow estimation, lane directions are automatically detected for daytime scenes. Foreground moving edges accumulated along the traffic flow direction are used as features. We utilize an adaptive clustering algorithm to estimate vehicle count for each frame. For nighttime scenes, statistical features are extracted from the segmented blocks, and regression models are applied to generate per-frame vehicle count. Finally, an effective graph-based mapping method is incorporated to map the vehicle count sequences to per-minute traffic flow. The accuracy of the traffic flow analysis is satisfying even when the cameras are seriously affected by rain. The experiments demonstrate that the proposed system can effectively analyze traffic flow under rainy conditions for highway surveillance cameras.