Design Of Welding Image Processing System Based On FPGA

Real-time detection of seam position is one of the key technologies of intelligent welding robot vision tracking system Since the processing of image data is relatively large,there is usually a lack of real-time processing problem in some applications when obtaining the weld seam position from the image. This paper presents a FPGA processor as the core of the line structured light seam image processing system to improve the extraction rate of feature point. That assists the PC obtain the location of the weld groove, depth and shape information more quickly.The system design has the function of line structured light weld image processing and image acquisition, including the acquisition, cache, processing,display of the line structured light weld image and Gigabit Ethernet data communication between FPGA and PC. The analog video signal from the camera of Line structured light vision sensor is converted by the ADV7180 to digital signal, then it achieves the SDRAM cache of a frame image which is realized by the FPGA chip. VGA control module achieves the image data reading from the SDRAM and the image processing. Communication part completes Gigabit Ethernet data communication based on rgmii interface and UDP/IP protocol. and maximum transmission rate ofeffective image data for 99MB/s. The time interval fom the selection of image data to the first valid data received by the host computer is 6.39 us.In the part of image processing, the advantages of the FPGA parallel computing ability are fully utilized, the center line extraction method of line laser stripe and the extraction method of weld feature points based on FPGA are studied. The center line extraction method of the laser stripe uses the regularity of the fringe data, largely avoids the influence of the reflection area of the structure light weld image. Seam feature point extraction used the method of fringe center point slope analysis and combined the least square method. The straight line fitting based on the least square method is more accurate for each laser stripe center line. Finally, the fitting line intersection point is calculated to obtain the coordinates of the characteristic points of the weld. After a lot of experiments, it shows that the extraction of the feature points of the weld has good stability, and feature point error meet the requirements of the control system for seam tracking. The delay time of image processing module in the image data transmission path is 178.9 us.In this paper, the extraction of the characteristic points reflects the higher real-time performance. The time from the CCD camera begin to output a frame image to the image feature point coordinates are sent to the host computer is in between 40.185ms to 57.965ms ,incudes the time of camera output a complete frame image which is 40ms.