Based On The Real-time Intelligent Video Analysis Algorithms Research And Application Of Cuda

Moving target detection and tracking is an important research branch of the intelligent visual surveillance, which provides a foundation for a higher level analysis. GPU computing that rises in recent years provides an effective method to improve the speed of digital image processing. This thesis mainly investigates GPU-accelerated digital image processing technology, in order to improve the real-time ability of the moving target detection and tracking system. The main contributions of this thesis are as follows:(1) This thesis studies the general-purpose GPU computing technology, and gives an introduction to NVIDIA’s CUDA parallel computing architecture, where CUDA enables the GPU to solve complex computational problems.(2) Three background subtraction algorithms are introduced, including single Gaussian model, Gaussian mixture model and ViBe algorithm, and their advantages and disadvantages are analyzed. Meanwhile, this thesis analyzes the characteristics of Gaussian mixture model and ViBe algorithm, giving the two algorithms’CUDA parallel solutions. Primary experiment results show that CUDA parallel implementation’s processing speed is several times faster than that of the CPU platform implementation.(3) Three shadow suppression methods are introduced to remove the moving shadows in foreground images. At the same time, the CUDA parallel solutions of these two shadow suppression methods, which are compared with the CPU-based implementations are discussed. The thesis is also concerned with the morphological filtering technology, and gives dilation’s and erosion’s CUDA parallel solutions. Both of their processing speeds are faster than those of Nvidia’s NPP library functions.(4) Data association algorithm focusing on the joint probability data association algorithm is investigated. Making use of background modeling technology, shadow suppression method and morphological filtering technology, the multi-target tracking algorithms based on motion detection and joint probability data association are obtained. Then tracking algorithm has been evaluated and analyzed. In addition, GPU-accelerated tracking algorithm is applied to obtain a certain speed improvement.