Real-time High-speed Visual Tracking System Development

Target tracking is one of key technologies in the field of machine vision. We can further realize many appliances such as movement analysis, behavior understanding, video compression though tracking moving target. Target tracking technique has been widely used in military, security, transportation, entertainment, robots and many other fields. A typical tracking system is usually based on platform cameras. In such a system, a camera is usually drived by actuators to rotate and to change the gaze direction, thus realizes the target tracking. This approach is limited by the mechanical and electrical transmission delay, the speed of adjusting gaze direction is not fast to match the image acquisition and processing speed, which result to not achieving high-speed track. To solve this problem, this paper develops a special real-time tracking system. The main work and achievements are as follows:(1) As the existing tracking system cannot realize high speed tracking because of platform structure and control constraints, a set of mechanical and optical path system is designed to change the way to adjust the gaze angle of the camera, which make sure the speed match of camera gaze angle adjustment and image capture and processing.(2) In view of the existing high-speed cameras are often very large, a small size high-speed camera is designed and experiment verifies that the camera can capture image stably and clearly with speed of 221FPS@640×480.(3) In view of the existing high-speed processing boards tend to be with large scale and insufficient performance, a high-performance image processing board based on FPGA + DSP is designed. We finished schematic design, PCB design and hardware board debugging. An experiment is designed to verify high-speed transmission capacity of the board.(4) In view of the difficulty that the tracking target in natural scenc is under change of appearance or cluster background, a kernel correlation tracking algorithm framework based on deep convolution neural network is developed. Large numbers of experiments are made to verify the accuracy and robustness of the algorithm framework. Besides another fast algorithm is proposed for high-speed moving target, and experiments verify that the fast track algorithm still have high tracking accuracy,which can adapt to the requirement of real-time high-speed tracking system well.(5) A simple computer program is developed for the tracking system, and a table tennis track experiment is conducted to verify the ability of high-speed image acquisition, transmission and track processing of the designed track system.