Research And Development Of Visual Tracking System Based On Embedded Processor

To identify and track items on the industrial assembly lines,a kind of visual tracking system based on embedded processor is researched and developed in this thesis.With the improvement of industrial level,the machine vision technology has been gradually introduced to the industrial production and used for identification and location of items,to increase the flexibility of production line.But it is mostly based on the PC or other professional platforms,which leads to high cost,low level of integration and also difficulties to popularization and application.However,embedded system has characteristics like high integration,low cost,high flexibility and adoption of hardware and software etc.It has much practical significance to the research and development the visual tracking system based on embedded processor.A complete visual tracking system based on embedded platform is designed and realized in this thesis,which takes the actual environmental conditions into consideration and referred to the existing research results and references.The visual tracking system includes the mechanical verification platform of,the motion control system of and the target recognition and tracking system.The main works of the thesis is as following:1.A mechanical verification platform to the proposed visual tracking system was designed in the thesis.The structural design and materialsselection to the platform was analyzed,also the hardware and software of the motion controller was designed according to the requirements.Using the technologies of machining and 3D printing,the experimental platform was constructed and realized.The repeating position accuracy was tested by experiments,which ensured that it could be used to the proposed system.2.An S-curve acceleration and deceleration algorithm is designed by combining DMA transport with timer cascade.The principle and implementation process of this algorithm are illustrated in this article.The algorithm is mainly used to reduce the vibration frequently occurred as the mechanical motion platform starts and stops.It is intended to not only lay a solid foundation for the following target recognition and tracking,but also improve the mobility of the motion platform.3.An improved MeanShift tracking algorithm with Kalman prediction functions is implemented.This algorithm is built on actual conditions and tracks all objects in a single frame image.First,you can prioritize tracking sequence under low computation load,then track the remaining targets through node state update and the MeanShift algorithm.4.A Raspberry Pi 3B platform is built.OpenCV and WiringPi of GPIO driver library are implemented.USB Webcam driver for Linux is analyzed in the thesis.A supportive system environment and development infrastructures are built for communication and target tracking byinstalling and analyzing the above systems and software.