Store entry rate statistical method based on yolo and centroid tracking

Abstract

The invention discloses a store entry rate statistical method based on yolo and centroid tracking, and relates to the field of monitoring video, deep learning, computer vision and pedestrian detection, and the method comprises the steps: carrying out the real-time detection of a picture in a monitoring video through a yolo target detection model; Therefore, obtaining the target category and the corresponding target position in each frame of image data, screening out the category belonging to a person, and the centroid position of the target is stored. And tracking the target through a centroidtracking algorithm, keeping the unique distribution of the target ID, calculating the target motion direction through the upper and lower frames of the video, and counting the number of people entering the store and passing through the store to obtain the store-in and store-out rate. The invention provides a store entry rate statistical method based on yolo and centroid tracking, and the method is simple in algorithm, is beneficial to software realization, and achieves the statistics of the movement direction of pedestrians through target detection and a centroid tracking algorithm.

Description

technical field [0001] The present invention relates to surveillance video, deep learning, computer vision, and pedestrian detection, in particular to a statistical method for store entry rate based on yolo and centroid tracking. Background technique [0002] Nowadays, there is a relatively bad situation. There are many people passing by the store every day, but very few people enter the store. However, a large passenger flow means high rent and high operating costs. If the high cost cannot be paid accordingly , It is undoubtedly the beginning of the decline of the store. In order to avoid this from happening, most stores will hold activities to attract customers, such as full gift activities, lottery draws, discounts, etc. As a store, it is necessary to understand which of these activities are the correct interpretation of pedestrian psychology, and install a The camera counts the traffic flow and the number of people entering the store, so that it can be concluded which a...

AI Technical Summary

Problems solved by technology

Manual counting is laborious and labor-intensive, and manual counting will have many disadvantages for large flow of people; it has limitations for sensor counting pedestrians, and now infrared is a more common sensor counting method, in order to achieve detection accuracy, only a single channel can be set , not suitable for a large number of pedestrian counts in public places; pedestrian counting based on computer vision can obtain passenger flow through real-time video analysis, which is universal and convenient

[0005] For pedestrian detection, in the existing methods, the histogram of orientation gradient is used as the descriptor of pedestrian detection, and then SVM is used for classification. -stage and two-stage have high precision, but the calculation speed is slow, and the effect of real-time target detection cannot be achieved

Images