Object Tracking Algorithm Optimization And FPGA Implementation Based On Correlation Filter

Object tracking is a branch of machine vision research that has a very important application value.The Object tracking algorithm can predict the target position in the video based on the existing initial information of the Object,which has very demanding requirements on accuracy and speed.However,in the actual algorithm research,the two attributes of high accuracy and fast speed are often not satisfied at the same time.Based on this,the performance of target tracking algorithms needs to be traded off,taking into account the efficiency of the algorithm and the structure of the implementation platform.The purpose of this paper is to study the optimization of the minimum mean square error sum-of-squares(MOSSE)based correlation filtering object tracking algorithm and the implementation of its target tracking system on a SoC FPGA platform.The MOSSE algorithm has the characteristic of fast tracking speed,which has a great advantage over other target tracking algorithms in the scenario of real-time target tracking.However,at the same time,MOSSE algorithm has the disadvantage of low accuracy.In complex tracking environments,the traditional MOSSE algorithm is often difficult to meet the accuracy requirements,and for this problem,the following research is made in this paper.The purpose of this paper is to study the optimization of MOSSE-based correlation filtering object tracking algorithm and the implementation of its object tracking system on SoC FPGA platform.MOSSE algorithm has the feature of fast tracking speed,which has great advantage over other object tracking algorithms in the scenario of real-time target tracking.However,at the same time,MOSSE algorithm has the disadvantage of low accuracy.In complex tracking environments,the traditional MOSSE algorithm is often difficult to meet the accuracy requirements,and for this problem,the following research is made in this paper.Next,the hardware implementation and improvement of MOSSE algorithm.The two-dimensional fast Fourier transform(2D-FFT)is one of the core computational steps of the MOSSE algorithm,and the optimization of resource consumption,power consumption and time delay of its hardware implementation is very important for the engineering application of the MOSSE algorithm.To this end,this paper designs a classical 2D-FFT structure based on the principle of 2D-FFT and improves it by using state transformation and data stitching methods.Through software analysis and comparison,an implementation structure that combines less resource consumption,lower power consumption and shorter delay is selected.Further,the MOSSE tracker is first decomposed into different stages for hardware description language design based on the idea of chunked design.This design is then simulated and verified using video sequences from the target tracking dataset.Then the overall implementation of the system is done on a SoC FPGA development board.Finally,the tracker is tested using complex data collected in a real-world environment to verify its tracking performance.