Abnormal behavior detection for harbour operator safety under complex video surveillance scenes

Abstract

In this paper, we analyze the problems faced by current video surveillance systems in the safety detection of port operators, and systematically research the methods of information extraction and operator behavior warnings. We collect a large number of port operation videos, and establish a large-scale port operations scene dataset named Harbor Dataset. We propose a deep learning based object detection algorithm to carry out the safety detection of the operator in the harbor scene. By comparing the detected personnel position and the location of the calibration area, we can judge the violation of the operator. We also use real-time tracking and data retention of offending operators to reduce the operation of illegal operators. Experiments show that our method yields a competitive result on Harbor Dataset in detecting the safety of the operating personnel and calibrating the dangerous operation areas.