Research On Customer Behavior Detection Algorithm For Supermarket Scene

In recent years,with the continuous improvement of people’s living standards and the continuous upgrade of consumption power,people have put forward higher and higher requirements for the retail industry.Under this general trend,the new retail industry represented by smart retail and unmanned supermarkets has developed rapidly.Compared with the traditional retail industry,the new retail industry relies more on intelligent solutions to obtain customers’ choices and preferences,and adjusts its own production and management strategies accordingly.This is inseparable from recording and analyzing the behavior of customers in supermarkets.Generally speaking,by detecting and recognizing the interaction between customers and products,such as touching,picking,etc.,you can determine the customer’s attention to different products to a certain extent.Customer behavior detection algorithms need to include at least two modules:human detection and behavior recognition.Existing algorithms often exhibit problems such as poor real-time performance and low generalization ability when deployed in actual scenarios.Therefore,it is of great significance to study customer behavior detection algorithms for supermarket scenarios.This paper mainly studies the customer behavior detection algorithm for supermarket scenes,proposes a 3D human key point detection algorithm based on binocular images,and a human motion recognition algorithm based on graph convolution network,aiming to solve the traditional human key points The speed and accuracy of the detection algorithm and the motion recognition algorithm are not up to standard in practical applications.This paper mainly conducts research from the following aspects:1.3D human key point detection algorithm based on binocular images.Aiming at the problem of poor real-time performance and low accuracy of the traditional human key point detection algorithm in the overhead view,several existing human key point detection algorithms are studied and compared,and an end-to-end 3D human key point detection algorithm based on binocular image is proposed.2.Convolution feature extraction network optimization algorithm.Aiming at the problem of poor expression ability of traditional feature extraction network architecture in 3D human key point detection tasks,a feature extraction network optimization algorithm based on attention mechanism and deep separable convolution is proposed,which improves the 3D human key point detection network Performance.3.Behavior recognition algorithm based on graph convolution network.Aiming at the problem of poor real-time behavior recognition algorithm based on RGB images,a behavior recognition algorithm based on graph convolution network is proposed,which improves the accuracy and real-time performance of the algorithm.