FPGA-based Grayscale Template Matching Algorithm And Its Application On Real-time Visual Object Tracking

Visual positioning and real-time tracking technology has always played an important role in the fields of scientific research and industrial applications.With the deep development of visual positioning and tracking algorithms,how to improve their accuracy and real-time effectively have always been a hot and key research point.At present,most of image processors are mainly serial processors based on the traditional x86 architecture,which hardly meet the requirements of high real-time application due to their low image processing speed and low efficiency.Because FPGA technology has the advantages of parallel and pipeline processing,it is more and more widely used in image processing.In this thesis,FPGA is adopted as an embedded image processor,and two gray-scale template matching algorithms using FPGA are developed.An embedded vision target positioning scheme based on the presented FPGA algorithm is proposed and applied in a target real-time tracking system with a Delta parallel robot.The main research works are as follows:1)Two kinds of FPGA hardware IP(intellectual property)cores for gray template matching algorithms are developed.An image distortion correction with bilinear interpolation is implemented on FPGAs,and a Matlab/FPGA hardware simulation environment is built up.Two kinds of grayscale template matching FPGA IP cores are developed and their detailed timing are defined,which are Sum of Absolute Differences(SAD)and Zero-based Normalized Cross Correlation(ZNCC).Compared with the PC-based gray template matching algorithm(using OpenCV),it is shown that our presented semi-parallel hardware algorithm has great advantages in processing time and resource consumption while maintaining good matching accuracy.2)Aided functional FPGA modules for visual tracking are designed.These modules include camera driver and image capture module,SDRAM controller module,VGA driver module,rectangular bounding box module,and serial communication module between FPGA and PC.The experimental results show that each module works well and can accurately realize serial transmission and reception of the target position information between the terminals of FPGA and PC.3)A real-time visual tracking experiment system is constructed.A special light path scheme with a semi-transparent film for optical filtering and 45° plane mirror is proposed,and a real-time visual object tracking platform using a Delta parallel robot is built.A motion control interface for visual tracking is programmed in a principal computer.The camera internal parameter calibration and hand-eye calibration between the camera and the robot are implemented,thus their transformation relationshipsis is determined.4)The real-time target tracking experiments based on our presented FPGA template matching IP cores are implemented.In order to verify the real-time tracking performance of the designed grayscale template matching FPGA IPs,,the experiments of target image matching accuracy,repetitive positioning accuracy of Delta parallel robot based on visual servo,real-time target trajectory tracking and dynamic target tracking performance evaluation of robot are implemented under the tracking experiment system.The experimental results verify the feasibility of the proposed hardware-level matching algorithm in the target tracking application by using a Delta parallel robot.