Headlights classification for traffic surveillance using a structure tensor method with PADI

Headlights classification is the most important part in traffic surveillance at night time. A main problem of those studies is streetlights and light reflections with similar shapes with headlights. Only one frame information may not be enough for the night-traffic surveillance. This paper proposes a method using previous frame information for classification. To classify headlights, a structure tensor is constructed using 3D Sobel operator masks in a given frame sequence. In addition, we use Positive Accumulative Difference Images (PADI) to remove non-moving bright spots in the scenes. The sample data is obtained by recording video in the night time. Experimental results show that the proposed method performs better than a traditional method about 10 % with regard to Recall, Precision, and F1 score.