A Robust Approach for Students Detection via Multi Cameras with Mask-RCNN
Qingxiang Wang, Wentao Li, Huaihui Liu, Liang Shan · 2021
With the development of science and technology and continuous improvement of social and economic level, the use of video surveillance equipment has got dramatically increase. It is an effective and robust method to execute people detecting and counting work that use continuous multi-videos streams directly and accomplish automatic people counting in matching analysis system. We use the improved Mask-RCNN via multi cameras for students detecting and counting. Students of overlap regions in multi-cameras also need re-calibration to compensate in statistic and analysis system. Excellent algorithms were proposed for students detecting in a multi cameras using environment. The developed multi camera approach achieves high object detection performance in our handcrafted dataset. Our multi-cameras students detecting and counting system, which train with smaller handcrafted dataset and align overlap region for counting accuracy improvement, gets a good accuracy in higher frame rate. This methodology is fill of challenging particularly testing in a higher real time video frame rate. The proposed method is expected to get excellent performance while it was extended to more complicated students recognition situation or relevant tasks.