Research On Algorithm Of Vechicle Shadow Detection And Removal

Shadow Detection, Intelligent Transportation system (ITS), Wavelet Transform, photometric invariant featuresIntelligent Transportation system (ITS) plays a more and more important role in the modern planning and management of urban traffic. The car shadows which move with the automobiles abounded in video detecting can disturb the ITS function, that is they lead the system mistaken the shadow for the car body, thus the difficulty of image segmentation, target localization and tracking which belong to the traditional calculating and visual method rises. For that reason, vehicle shadow should be detected and removed primarily.In this paper, Phong Illumination Model is applied to analysis the mechanism of shadow formation in road traffic image and the image characteristic of shadow region, after the analyzing, results come out that the brightness of covered area is decreased, however the textural features remains invariable, either does the photometric features of the shadow region, no matter it is covered or not. The article starts from the image characteristic of shadow region, basing on the treatment of'quotient- proportion'method to the realistic terrain scene and background picture, then the vehicle shadow detecting 2-step way is derived as follows:1. Performing the wavelet transformation to the foreground picture gained from the'quotient- proportion'method, analyzing the wavelet coefficients of high-frequency and low-frequency separately , extracting and marking the coefficients which is suitable for the shadow characteristic, then operating the reconfiguration, until the quasi shadow region graph is acquired.2. Analyzing the current graph and background picture which is relative to the quasi shadow region graph, if criterion of photometric invariant is satisfied, the region can be seen as the true shadow area, vice verse.Basing on the shadow detecting, the running mechanism about how the vehicle shadow effects the electrical police system, 3 schemes of shadow removal calculation are given, And the applicable condition which is suitable for each one is discussed respectively. They are:1. The Algorithm of shadow removal based on the background; 2. The Algorithm of shadow removal based on the neighbor region;3. The Algorithm of shadow removal based on the Retinex theory.The experiment shows that the method used here has no special requirement to the color and size of the car or the projective direction of shadow, the average detected rate is over 92%, the average false - positive rate is below 3%, and the average running time is less than 30 ms. The shadow of bicycles and persons that run in the same graph with the automobiles are rejected and removed efficiently, on the other hand, better quality is reflected on applicability, efficiency and velocity processing.