Rear lamp based vehicle detection and tracking for complex traffic conditions

In video surveillance system, detection and tracking of vehicles are two foundational and significant tasks. In this paper, a vehicle detection and tracking method based on rear lamp pairs is proposed. The proposed method combines color with motion information to perform vehicle detection. In order to adapt to different weather conditions like night, the rear lamps are divided into two categories: unlit lamps and lighted lamps. First, threshold segmentation are used to extract both the unlit and lighted lamp candidates in hue-saturation-value (HSV) color space and the thresholds are selected automatically by maximally stable extremal region (MSER) method. Then, all lamp candidates are tracked by using Kalman filter and lamp candidates with short-lived trajectories are removed to avoid disturbances. Next, two adjacent lamp candidates with similar speed are bound together as a region of interest (ROI), which represents a potential pair of lamps. Image cross-correlation symmetry analysis based on Gabor filter is utilized to find the ROIs with symmetrical texture and these symmetric ROIs can be regarded as pairs of lamps. The experimental results show that the proposed method can effectively deal with various illumination conditions and improve the accuracy and robustness of vehicle detection. In addition, this method can perform vehicle detection and tracking under complex traffic conditions and the Gabor filter based symmetry analysis can successfully suppress subtle difference between the left and right parts of a vehicle as well as environment noises.