Online Image Preprocessing System Of Vehicle Wheel Sets Image Based On GPU

Recently,with the rapid development of railway transportation,the detection of the wheel sets,which is the important parts of the vehicle,has an important significance for protecting the safety of railway transportation.It's one of the trend of railway traffic detection technology to improve the vehicle wheel sets abrasion detection speed.Wheel sets image preprocessing is an important part of the detection of the vehicle wheel sets wear and it's benefit for improving the speed of the vehicle wheel sets wear detection to improve the speed of wheel sets image preprocessing.To improve the image preprocessing speed,we come up with the study of wheel sets image preprocessing technology based on GPU.The technology of wheel sets image preprocessing based on GPU is aimed to get continuous single-pixel outline curve by preprocessing the original wheel sets image obtained through the CCD camera.The continuous single-pixel outline curve can be used to calculate wheel sets wear parameters.The main contents of this paper were as follows.(1)Studied and summarized the existing wheel sets preprocessing techniques and analyzed their respective advantages and disadvantages.Proposed the image preprocessing method based on GPU to overcome the low efficiency.(2)Researched the parallel computing technology and development status and selected CUDA as development platform of GPU.Researched the hardware architecture of GPU and the programming model of CUDA and built the software and hardware platform based on GPU for implementing the image preprocessing parallel algorithm.(3)Implemented one of the image preprocessing algorithms,the mirror algorithm in CUDA parallel architecture and analyzed the performance of the parallel mirror algorithm.Optimization the parallel process by changing the execution configuration based on the performance analysis results.(4)Improved the segmentation method of the image preprocessing algorithms,and the region growing segmentation algorithm is replaced by a fast wheel sets gray image segmentation algorithm which is based on the CUDA parallel computing architecture and k-means and STING clustering algorithm.The algorithm based on CUDA parallel computing architecture,guaranteed the effect of image processing of the wheel sets,and meanwhile greatly improved the speed of image processing.Finally,analyzed the performance of the parallel segmentation algorithm.(5)Analyzed the parallelism of the hit-miss thinning algorithm of the image preprocessing and concluded that it's a kind of serial thinning algorithm.Then,found a kind of parallel thinning method which is suitable for wheel sets image by the comparison of the existing parallel thinningalgorithm,and made improvement on it and implemented on the CUDA parallel computing architecture.Finally,analyzed the performance of the parallel thinning algorithm.(6)Analyzed the parallelism of the pixel tracking algorithm of the image preprocessing and proposed parallel program and implemented it on the CUDA parallel computing architecture.Finally,analyzed the performance of the parallel pixel tracking algorithm.(7)Compared and analyzed the experiments results of image preprocessing based on CPU and GPU,and the experimental results show that the GPU had advantage in wheel sets image preprocessing.