Target Tracking Algorithm And FPGA Implementation Verification

In the field of computer vision,target tracking is a basic and hot research topic.Its application range is very wide and active,including video surveillance,traffic monitoring,UAV system,medical diagnosis and missile guidance,etc.Although the research of target tracking has made great progress in the past decades,due to the unpredictable appearance changes,such as partial occlusion,illumination change,geometric deformation,background clutter,fast motion and so on,the research of this subject still faces many difficulties and challenges.A typical visual tracking process starts with the initial position of the target in the first frame of the video(such as a rectangular tracking box),and then estimates the position of the specified target in the subsequent frames.In the existing tracking research,the sub block based method has become a research hotspot in recent years,which benefits from its robustness to local apparent changes,especially partial occlusion.By dividing the candidate region of tracking target into several sub blocks,the sub block based method attempts to extract some useful clues from the identified reliable sub blocks.However,in the process of capturing the large spatial dependence among sub blocks,there may be some difficulties,especially for the target with a large homogeneous region.In order to solve the above problems,this paper proposes a tracking method,that is,the multi-directional recurrent neural network(RNN)is used to track the related targets.The multi-directional recurrent neural network(RNN)is used to encode all the sub blocks from four different angles.The multi-directional RNN has high robustness in the task of target tracking.Considering the practical application,this paper also considers the target tracking on FPGA.The target tracking algorithm on FPGA must be reliable,real-time,meet certain tracking accuracy,and have certain operability.The proposed target tracking algorithm combines Harris method and optical flow method.Firstly,Harris method is used to extract some target angle features,and then optical flow method is used to match angle features more accurately for subsequent video frames.When the target is rotated or twisted,the center of gravity algorithm is used to calculate the center of gravity of the matching features.In order to meet the requirements of real-time tracking,a small area image search method and high-speed digital signal processing system are also designed.In addition,this paper also develops the object tracking algorithm based on particle swarm optimization parallel resampling on the stream based architecture.Resampling is completed in the effective pixel region of the input image,and particle prediction and update are performed in a synchronized region.Therefore,the proposed method can achieve 60 FPS real-time performance of VGA image without using any external storage device,and synchronize with the pixel throughput of the camera.The experimental results verify the effectiveness of the proposed method,which can provide reference value for some engineering applications.