Design Of MBD-based Configurable Object Tracking Algorithm Hardware Acceleration Architecture

Video object tracking technology has been one of the research hotspots in the field of computer vision due to its wide application value.Although the current object tracking algorithm has achieved good performance on many data sets and has been applied in actual scenarios,for specific application scenarios,it is still necessary to adjust the tracking algorithm and hardware structure according to the complexity of the algorithm and the computing power of the platform.In real-time tracking applications,FPGA accelerated designs are usually utilized to process complex algorithms at high speed.However,the FPGA development is too difficult and the iteration cycle from algorithm modification to FPGA verification is too long by using traditional FPGA development.To address the above issues,this paper proposes a configurable object tracking algorithm hardware architecture based on model design,which can quickly modify the tracking algorithm according to the characteristics of the scene,reasonaly configure the rellevant hardware modules,and form a specific object tracking algorithm,so as to achieve the purpose of efficient iterative and real-time tracking.The main research work of this paper is as follows:(1)Aiming at the problem that a single object tracking algorithm cannot be applied to all application scenarios,this paper proposes a configurable object tracking algorithm framework for MBD based on correlation filtering.According to the characteristics of the scene,the framework can reasonably configure related algorithm modules,that is,feature extraction module,filter design module,scale processing module and model update module to form a specific object tracking algorithm.Furthermore,this paper optimizes the above four modules to adapt them to FPGA implementation.(2)Aiming at the problem that the traditional FPGA method cannot quickly implement the object tracking algorithm and iterative optimization,this paper proposes a hardware acceleration architecture of a configurable video object tracking algorithm based on MBD.Efficient parallel pipeline processing is implemented between and within the modules of the architecture,which reduces the occupation of resource space and access to storage space.In addition,this paper utilizes Simulink platform to design and implement the hardware acceleration architecture from top to bottom,and utilizes hardware simulation tools to analyze key signals.Finally,this paper compares the designed architecture with the results of PC processing by visual method,the results demonstrate that the architecture can achieve the accurate positioning of the interested objects.(3)This paper utilizes Fixed-Point tools in Simulink to perform fixed-point operations of the configurable object tracking hardware acceleration architecture.In addition,the relevant tools in HDL Coder are utilized to automatically perform code convert and optimize the timing and area of the critical paths,so as to ensure the better performance in terms of resource occupancy and throughput.After the structure adjustment and optimization,this architecture is deployed on the FPGA of Xilinx's ZYNQ ultra-scale MPSo C series ZCU106,compared to the corresponding algorithms on the PC,it is found that the architecture proposed in this paper can meet the real-time requirements of the object tracking algorithm.