People Detection and Finding Attractive Areas by the use of Movement Detection Analysis and Deep Learning Approach
Ebrahim Najafi Kajabad, Sergey V. Ivanov · Procedia Computer Science · 2019
This paper presents a technical approach related to the video computer analysis, to detect people and control the behaviour of people. Control the behaviour of people in public place can be a benefit for understanding the share of overall traffic your area is attracting. Find out what, encourage customers to buy products are most crucial for big companies to increase the sales rate and to improve the quality of customer service. We use surveillance cameras, which located in the museum. We offered two methods, the first method for detecting people in a closed space and second method finding density areas which people more spend time to visit. The YOLO model makes predictions with a single network evaluation. Systems like R-CNN and Faster R-CNN, on the other hand, make multiple assessments for a single image, making YOLO extremely fast, running in real-time with a capable GPU. For detect people used YOLOv3 algorithm which is published by [18] and shows that it has high accuracy to identify people, also we compared the proposed method with other detectors, HOG, SSD and YOLO-tiny which shows the proposed algorithm has better performance in this point. And for finding density areas, We utilized a background subtraction with Gaussian Mixture algorithms and heatmap colour technique to analysis each frame and figure out, where are the density areas which shows people like to spend more time to visit. The experimental results have shown that the accuracy and the performance of both algorithms are quite good.