PC-PINet: Partial Re-identification Network for People Counting with Overlapping Cameras
Laihui Ding, Shengke Wang, Rui Li, Long Fei Chen, Junyu Dong · 2021
People counting is one of the important tasks in intelligent video surveillance. However, reliable people counting in the overlapping areas under multiple cameras is still a challenging task due to the occlusion, small object, and minor individual similarity. In this paper, we propose a novel partial body detector to detect the upper bodies of the people in the images captured by two cameras, and our proposed partial body detector can achieve the balance between the speed and the detection accuracy. Furthermore, we propose a Convolutional Neural Network based person re-identification framework, namely Partial Reidentification Network (PINet), which make full use of the multilevel features and the identity information. To demonstrate the effectiveness of our propose method, we collect classroom datasets and conduct extensive experiments on the dataset. The experimental results show that our proposed method achieves excellent performance on the classroom dataset.