Research And Implementation Of Real-Time Target Tracking Hardware Processing System

Target tracking technology is widely used in industrial pipeline monitoring,pod tracking,public safety and other fields.As a processing technology based on image understanding and video analysis,the target tracking technology is of great significance for machine intelligence research.At present,the tracking accuracy and robustness of the target tracking algorithm have been improved.The target tracking algorithm based on PC has the disadvantage of slow running speed.FPGA processors that are used in hardware tracking systems can be used for algorithm acceleration by relying on their own parallel operation and high-speed pipeline mechanism,but it is difficult to deal with more logical discriminant class processing in complex tracking algorithms.To this end,this paper combines the advantages of ARM processor to compensate for the disadvantages of FPGA processor in the complete implementation of tracking algorithm,and develops a multi-function target tracking hardware processing system,which can be used with the servo control system to complete the offline work of image data stream collection,information processing,high-speed transmission and real-time tracking display.The content covers:(1)The development of target tracking algorithm and its implementation status on different hardware platforms are reviewed.The dual-core processor architecture based on FPGA+ARM is determined.In the hardware tracking system,the FPGA processor has been widely used due to its high degree of parallelism,but it is not suitable for performing tasks which include more logic judgment in tracking algorithm.The ARM processor which has the advantage of being good at logical discriminant operation can just make up for this deficiency.In order to ensure that the system has a hardware foundation that can be accelerated in parallel,and has strong logic processing capability,achieving high-speed and robust design goals,this paper uses the hardware tracking system architecture based on FPGA+ARM dual-core processor.(2)According to the determined hardware architecture,in order to make the system have the advantages of small size and easy expansion,the processor backplane and the top board circuit which integrates the image acquisition and display interface are designed respectively.The development and verification of the USB3.0 high-bandwidth communication circuit which is based on the FX3 peripheral controller between the processors have also been finished.The one-way communication speed is nearly 320MB/s,and the volume of the whole functional circuit is 112mm×107mm×30mm approximately.(3)The normalized cross-correlation matching tracking algorithm which is suitable for implementation on FPGA processor is studied.To solve the problem of strict template requirements,a template updating strategy combining the kernelized correlation filters algorithm is proposed.Through the relevant MATLAB simulation experiments,the rationality of the system's 8×8 matching template size and the strong anti-jamming performance under complex background are verified.By comparison,the scaling process and template update mechanism proposed in this paper can improve the running speed of the original normalized cross-correlation algorithm by about 35%.For the commonly used target linear motion scene,the average accuracy measured is as high as 94.67%.(4)The related module implementation and simulation debugging tasks of the system algorithm on the FPGA processor are completed based on Verilog language.A double-three interpolation real-time computing architecture based on lookup table is designed.When the working clock frequency of the source image data stream and the module are 25 MHz and 100 MHz,the field blanking and line blanking interval are about 91 us and 10 us respectively.The maximum magnification can be about 4 times.In addition,the normalized cross-correlation formula for hardware calculation is derived.Based on this,an innovative architecture of 16 multiplier parallel operations is proposed to facilitate real-time pipeline operation.Through practical test,the tracking system developed in this paper is proved to have the interactive functions of image adaptive zooming with adjustable precision and first frame target selection.For a target with motion speed of less than 0.75 W pixel/frame(W is the target size)in a visible light camera image,which has a resolution of 752×480 and a frame rate of 30 Hz or even higher,the system can track and display in real time,highlighting the characteristics of fast and efficient,which has good theoretical and practical application value.