Research And Implementation Of GPU-accelerated Vehicle Detection And Tracking

With the development of intelligent transportation system and intelligent vehicle, drive-assist in vehicle system has become a research hotspot. Lane detection, vehicle detection and tracking involve a lot of complex image processing calculations, but the characteristic of serial processing in CPU can’t meet the real-time requirement in practical application. However, graphics processing unit (GPU), which has become a kind of standard device in vehicle electronic system, has powerful parallel computing capacity to solve this bottleneck. In order to meet the real-time requirement in practical application for vehicle detection and tracking, the paper studied GPU-based methods for acceleration in depth, which are as follows:1A method of GPU-accelerated barrel distortion correction was designed. The parallel computing resources of GPU was used to accelerate the calculations of pixel coordinates and color values in correction image.2A method of GPU-accelerated lane detection based on Hough transform was designed. The parallel computing resources of GPU was used to accelerate the calculations of graying, binarization, median filtering, Sobel operator for edge detection and Hough transform.3A method of GPU-accelerated Harris corner detection was designed. The parallel computing resources of GPU was used to accelerate the calculations of graying, initial smoothing, Gaussian filtering, interest value and removing false corners.4A method of GPU-accelerated mean shift for vehicle tracking was designed. The parallel computing resources of GPU was used to accelerate the calculations of distance weight matrix and color values in target area, color values in candidate area and offset vector matrix.5A system for GPU-accelerated vehicle detection and tracking on the highway was implemented. The co-processing mode of GPU+CPU was used to implement detection and tracking of vehicle which was in front of a moving vehicle. The calculations of lane detection, vehicle detection, and vehicle tracking were accelerated by GPU.In this paper, GPU-accelerated experiments were carried out by using C/C++, GLSL and OpenGL2.0on the platform of Visual Studio2008in Windows XP system. The results showed that, compared with the implementation on CPU under the same conditions, the implementation of GPU-accelerated vehicle detection and tracking had better real-time performance, which achieved the desired goal and could be utilized in practical applications.