Intrusion Detection and Tracking at Railway Crossing
Naxin Cai, Houjin Chen, Yanfeng Li, Yahui Peng · 2019
Railway crossing safety is an issue of great public concern. With the development of computer vision, intelligent video surveillance was widely used for object detection. However, most existing object detection methods could not perform well in outdoor environment, especially under the bad weather conditions. In this paper, an intrusion detecting algorithm was proposed for railway crossing. Our method consists of the following three steps. First, three nonparametric background models with different learning rates are designed for the detection of moving and static objects. Second, an object tracking strategy is introduced for reducing false positives in the detection. Finally, a feedback mechanism is introduced to selectively update the background models when static objects are removed. Our method was tested on real railway sequences and the public i-LIDS datasets. Experimental results showed the proposed method achieves accurate detection for both moving and static objects and some false positives caused by rains and partial occlusion could be reduced.