Vehicles Shadow Detection Based On RGB Color Space

As science and technology develop rapidly, the number of vehicles increasesdramatically, this leads to congested roads, traffic accidents frequently. In order toeliminate the negative effect of vehicles, the experts and scholars continue exploringmethods. Intelligent Transportation Systems (ITS), as a new technology area, plays animportant role to solve traffic congestion, improve traffic efficiency.Shadow, which is a very common physical phenomenon, appears in traffic videoinevitably. The existence of the vehicle shadow may cause adhesions and vehicle shapedistortions, which brings adverse effect to machine vision, such as vehicle identification,vehicle retrieval and tracking. Shadow Removal is particularly important for solving otherproblems. Shadow detection is the key to shadow removal. The accuracy of shadowdetection algorithm is will affect the subsequent performance of shadow removal directly.In this paper, on the basis of the summary and analysis of the research at home and abroad,we analyze the cause of shadow, fully consider the image color space information, andpropose a novel shadow detection algorithm. Firstly, Gaussian mixture model (GMM) isadopted for foreground detection, and then each component of the RGB is analyzed in theshadow area. Color ratios like RGB ratios and the blue and red ratio (B/R) are combinedfor the initial detection of shadow. At last, morphological methods are utilized for spatialadjustment of initial detected results to get accurate detection of the shadow region. We useMATLAB to simulate the algorithm on multiple databases, and design of the three sets ofcomparative experiments to verify the performance of the algorithm. Including the contrastof whether combine the characteristics or not, the binding mode characteristics andcomparison with the existing shadow detection algorithms. Experimental resultsdemonstrate that the proposed algorithm is simple and effective, able to accurately detectthe vehicle shadow and suitable for real traffic scene, and the detection performance isbetter than some of the existing shadow detection algorithm.