Research And Application Of Mobile Robot Miniature High-speed Stereo Vision Technology

Stereo vision technology is a research hotspot in the field of robot vision and has a profound impact on the development of robots.The vision application of the robot requires its vision system to have real-time high-speed processing,high-precision measurement,and high robustness.It can adapt to different lighting scenes,overcome over and under exposure problems caused by changes in lighting conditions,and have small size and power low.This paper proposes a robot-based high-speed binocular vision technology based on FPGA,which can measure three-dimensional depth information in real time,high speed,and high accuracy.It has adaptive gain and exposure functions,and can be well adapted to robot vision tasks.The specific research work is as follows:(1)This article has developed a set of robotic high-speed binocular vision hardware system based on Xilinx FPGA,which can synchronously capture and process two color images,and use FPGA to accelerate image processing and stereo matching at high speed in real time.The two cameras of the system use hardware synchronization for exposure,and use the global exposure mode,which can achieve a maximum acquisition rate of 2592×2048 resolution 100 frames per second,which meets the needs of robot high-speed vision tasks.The hardware system is compact in structure,small in size,and low in power consumption,and can be installed on a mobile robot platform and used online.The hardware is divided into two independent image acquisition units and an FPGA image processing unit.The baseline distance can be adjusted according to the application.(2)This paper designs a high-speed binocular vision firmware system for the robot high-speed binocular vision hardware system developed.The firmware system includes image acquisition,image pre-processing,automatic gain and exposure control,stereo vision matching,and image USB output.The firmware system uses a combination of FPGA acceleration and an ARM processor.It uses FPGA logic to accelerate image processing and stereo vision matching tasks,and uses ARM processors to complete system control and USB protocol tasks.In response to the over-and under-exposure problems caused by changes in lighting conditions in robot vision applications,automatic gain and exposure control can simultaneously adjust the gain and exposure time of the two image acquisition units in real time according to changes in ambient light,making the system adapt to different lighting Scenes,effectively overcoming and underexposing caused by changes in lighting conditions.The entire firmware system can achieve real-time processing and output performance of 100 FPS at a resolution of 640×480.(3)This paper proposes a memory-optimized SGM stereo matching algorithm(MOSGM)based on FPGA,which meets the high-speed and high-precision requirements of robotic stereo vision tasks.This method uses an FPGA full-pipeline architecture and does not require the use of external DDR memory.This method achieves higher accuracy while using an optimized processing structure,which reduces the consumption of programmable logic resources in FPGA,and can achieve the performance of 640×480 resolution @ 325 fps at a 196-level parallax,which meets the robot binocular high-speed vision Task application requirements have higher use value.This paper combines the binocular high-speed stereo vision hardware,firmware and algorithms of the robot to achieve real-time,high-speed,high-precision,high-robustness stereo vision tasks.Experiments show that the depth measurement accuracy of this system is high,and the highest accuracy can reach 1mm,which has high application value.