Multi-View Convolutional Neural Networks Crowd Counting Model Based on YOLOX

Crowd counting refers to estimating the number of crowds and crowd distribution in images or videos, which can effectively manage pedestrian volume and observe the degree of crowd congestion in time. Single-view crowd counting has received a lot of attention in recent years and achieved remarkable performance on many public datasets. However, it is not suitable for wide-area occluded scenes due to field-of-view limitations. Multi-view crowd counting sets up multiple cameras in the same scene from multiple angles to complete crowd counting task. This paper proposes a multi-view convolutional neural networks crowd counting model based on YOLOX. Experiments are conducted on two public datasets (PETS2009, CityStreet). Results show that this method can achieve good counting accuracy and fast training speed.