Objects Detection And Tracking Technologies In Traffic Surveillance System Under Complex Weather

As digital cameras and powerful computers become wide spread, the number of applications using vision techniques has increased enormously. One such application that has received significant attention from the computer vision is traffic surveillance system. But normal outdoor surveillance system depends heavily on the illumination of environment, and may sensitive when the natural noises occur. We propose a new traffic surveillance system that works under the complex weather such as rainy, foggy and snowy. We treat the rain or the snow as a sort of image noise, and foggy weather as the contrast reduction. Then the propose surveillance system is equipped a powerful noise eliminate and contrast enhancement function blocks.By eliminate the image noises, we use a novel curvelet transform algorithm that has an adaptive curvelets shrinkage thresholds. As a new research hotspot, curvelet transform is a new multi scale representation most suitable for objects with curves. It can handle anisotropy edges and give a more sparse represents of curves using only a small numbers of coefficients. The curvelets are more powerful than wavelets for image denoising and contrast enhancement. In our system, we improve the threshold method of the traditional curvelets WRAPPING algorithm. The curvelets coefficients of an image are thresholded and shrunk within a small neighborhood window. This method can filter the noises in a such great performance that it can raise the peak signal-to-noise ratio (PSNR) for about 8-10 dB compared with the hard or soft threshold wavelet filters. The experiment results show that both the PSNR and vision of the curvelet denoised images improve a lot, and the curvelets can reconstruct the edges of the images well.The second highlight of this paper is developed a wavelet edge detect and shadow removal algorithm based on illuminance information for foreground objects segmentation. Shadow removal is a well known difficult problem in computer vision. In our method, we first use a 2-D Haar wavelet transform on the region of interest (ROI), then transform ROI region from RGB space to HSI space. The shadow are the region that with fewer edges, which means the maximal modulo of the wavelet coefficients are small, and with the lower illuminance. The experiment results show that our method can segment shadow of vehicle efficiently.After segmentation, the object tracking is another important research problem. In our system, we use a triple states Kalman filter to tracking the vehicle. By define the MCD (Maximum Close Distance) templets and relation matching rules, the tracking system can overcome the occlusion and miss tracking problem.