Real-time Crowd Detection and Counting in Video Sequences Using R-HOG Net

Abstract

Human detection and tracking helps in better law enforcement, video supervision and traffic management. It is a challenging errand because of the substantial impediments, observe variations, frequency of the people and appearance of the body in the video sequences recently proved that the performance of CNN based approaches for crowd counting are stronger compared to the conventional methods. This paper proposes R-HOG Net, a hybrid method that uses ResNet50 and Histogram of Gradient Descriptor to count people in huge gatherings, rallies etc., in a real time as well as from a pre-recorded video.The proposed system detects and tracks the faces and also the shape of human beings in different views. The system automatically updates the crowd count whenever a new face is detected. The performance of the proposed method is tested for various video sequences and found to be appealing by overcoming the limitations of Histogram of Oriented Gradient (HOG) and ResNet50 CNN structure.