Small Object Detection for Intelligent Abnormal Event Detection in Power Distribution Networks
Yize Tang, Zhang Yi, Wenjie Kong, Lanxin Qiu, Yuxiang Lv, Hao Wu · 2022 15th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI) · 2022
With the development of intelligent object detection technology, the traditional artificial monitoring system has been gradually replaced by the automatic monitoring system based on intelligent video analysis. In the daily operation of distribution network monitoring system, the monitoring of abnormal events associated with patrol inspectors is an important part of video analysis. Safety helmet compliance and the smoking behavior are key inspection points in these abnormal events in intelligent power distribution network monitor scenario. Due to small size of helmet and cigarette, it is hard to detect these small objects under standard object detection framework. To solve this problem, we designed a two-stage neural network framework to detect small objects attached on human body. At the first stage, we extract human body part from original image. Then, we use standard object detection framework to detect helmet and cigarette. With increased proportion of small object in the image, the proposed framework can detect small objects accurately and effectively. Experiment results and field test both demonstrated its effectiveness of the proposed algorithm.