Hardware-oriented Lightweight Object Tracking Algorithm Design And FPGA Implementation

Visual object tracking is an important research direction in computer vision,and it is applied in a wide variety of contexts,such as video surveillance,human-computer interaction,and autonomous driving.Traditional general-purpose computer-based visual object tracking systems have high power consumption and poor real-time performance,which is difficult to meet the needs of embedded systems for high-speed real-time object tracking.Compared with the traditional visual object tracking system,the visual object tracking Very Large Scale Integration(VLSI)system has the advantages of high speed and high integration,and can be widely used in various embedded visual object tracking systems.The purpose of this thesis is to design a visual object tracking VLSI system for edge computing devices or embedded systems to achieve low-cost,real-time object tracking.Traditional object tracking algorithms are not developed for hardware implementation.Visual object tracking VLSI systems designed based on these algorithms have problems such as high resource consumption and poor real-time performance.The thesis carries out collaborative research from the three aspects of visual object tracking algorithm,VLSI system architecture,and circuit design to solve the above problems.A hardware-oriented lightweight visual object tracking algorithm is proposed,and the architecture and circuit of a high-speed lightweight object tracking VLSI system are designed accordingly.The FPGA prototype of the designed object tracking VLSI system is realized,and the test platform is built to complete the test and analysis.The innovations of this thesis are as follows: 1)For visual object tracking VLSI system based on traditional object tracking algorithm design,which has high computational complexity and poor real-time performance,a lightweight object tracking algorithm is proposed,which is based on the object representation of unified textural and dynamic compressive features.The algorithm is oriented to hardware design and development,which has the characteristics of strong robustness and simple calculation.2)For the low resource consumption requirements of embedded applications,a visual target tracking VLSI system architecture based on the Memory-Centric mechanism is designed,and through the optimized design of the filtering sequence,the storage resource consumption is reduced by about 39%.3)Aiming at the demand of real-time object tracking,a multi-level pipeline circuit design is adopted.Feature extraction and feature matching circuits are designed with pixel-level,feature-level,and sample-level multilevel parallel processing to ensure the real-time performance of the system.After testing,the visual object tracking VLSI system designed in the thesis achieves low-cost and high-speed object tracking,and obtains robust tracking results.The VLSI system can achieve 600 frames per second at 100 MHz system clock and typical 320×240image resolution input.