Research Design Of Package Detection Tracking Counting System Based On Deep Learning

The 2018 China express development index report shows the rapid development of the express delivery industry,which is manifested in the large volume and high frequency of parcel transportation.With the industrial upgrading of express delivery industry,intelligent warehouse management is the development trend.The important part of the intelligent warehouse is the package information management system,which can be realized by modern information technology and package detection and tracking technology.At present,the existing package tracking and detection systems are based on RFID technology and laser line scanning,both of which have obvious defects.In this paper,a package detection and tracking system based on deep learning is designed.The function of this system is to obtain the package traffic data of each monitoring point,and then upload these important data to the scheduling system of the warehouse to realize the intelligent management of the warehouse..The main contents of this paper are outlined as follows:(1)The hardware and software schemes of the package detection tracking counting system based on deep learning are designed.The hardware system mainly includes the sensor part,the computation part and the display part.Software system includes software module design,software flow design and so on.(2)The model structure and network structure of YOLO series target detection are analyzed in detail.The improved method of YOLOv3 model was analyzed and studied,and the model structure was changed to increase the detection accuracy of the model.In addition,the compression principle of the model and the commonly used compression methods are analyzed,and the compression of the complex model is realized.(3)The principle of classical Kalman filter and DSST correlation filter is analyzed in detail,and an improved target tracking scheme based on Kalman filter and DSST correlation filter is realized.In addition,a target tracking scheme based on measurement is designed,and the effects of three target tracking schemes are compared.(4)The interface design and realization of this system are introduced in detail.Then,the module parameter setting and identification region selection before system operation are introduced.Finally,the system module is tested,the whole system is tested offline,and the whole system is tested online.This thesis mainly aims at tracking,detecting and counting the packages to complete the software and hardware design of the system.With deep learning target detection technology and target tracking technology as technical support,the system software was written by C++ in the Windows development environment.Finally,the system was tested at a monitoring point of a company's warehouse sorting line.