Motorbike theft detection based on object detection and human activity recognition

Motorbike theft detection from surveillance videos is not only a challenging problem of object detection and human activity recognition in the field of computer vision, but also an urgent need for preventing theft crimes in real life. In this paper, we propose a framework for motorbike theft detection based on the combination of object detection and human activity recognition. In order to reduce the number of objects that are needed to be processed; we estimate the regions of interest in videos and only evaluate objects in these regions. We then analyze the activity sequences of thieves from video clips and use this result for theft detection. The system will sound an alarm if the activity sequences recognized from the video match with ones of thieves. In addition, we build a motorbike theft dataset for evaluating the performance of our framework. Experimental results show that our proposed framework works well on the reality dataset; it proves to be a feasible and applicable solution.