Research And Implementation Of Customer Identification Technology For Pedestrian Counting

As an indicator of offline business operation,human flow has always been focused on by the managers of shops.At the same time,human flow plays an important role in the operation and decision of shops.Since the beginning of 21 st century,the managers of shops began to explore the potential links between human flow and daily turnover.However,since the statistics for human flow is a long and persistent task,the process of acquiring human flow has problems of inaccurate acquisition and large amount of manpower consumption.Even in this way,with the concept of customer portrait proposed in recent years,the requirement for multi-dimensional information collection and simple process for obtaining human flow cannot satisfy the decision requirement of shops in complex business environment.There are fatal defects in the statistics of human flow using both artificial statistics and current large-scale use of WiFi probes for automatic human flow.The artificial statistics which need much manpower is certainly not desirable.However,since the WiFi probes need customers to carry the Android WiFi probe and need "mobile Internet" service,it is doomed that the human flow statistics obtained using this method is not accurate,let alone using the wrong data to guide the daily decisions of shops.With the development of deep learning algorithm,the task of acquiring human flow statistics has a new solution.In view of the characteristics of human flow statistics and combing with existing deep learning algorithm,this thesis proposes a new pedestrian re-recognition algorithm based on multi feature metric.By using the proposed algorithm,high accuracy of human flow statistics can be achieved.The main work of this thesis is listed as following:(1).We study the background and recent developments of human flow statistics.In the meanwhile,we also study the detection algorithms used in the data acquisition of human and human face in order to obtain accurate data source.(2).We propose a pedestrian re-recognition algorithm based on multi feature measurement.(3).We compare the proposed algorithm with the prevailing pedestrian re-recognition algorithm.The effectiveness of the proposed algorithm is then validated.(4).We apply the proposed algorithm to the human flow statistics of the offline store.The effectiveness of the proposed algorithm is again validated by the successful acquisition of the daily human flow.