Increasing the Accuracy of Detection and Recognition in Visual Surveillance

Visual surveillance has two major steps of detecting and recognizing moving objects. In the detection stage, moving objects must be detected as quickly and accurately as possible and the influence of environmental light changes and waving trees should be reduced. In this research a block-based method is introduced in HSV color space in the detection stage. This method did not scan all the pixels of the frame and acted well in situations like sudden light changes. A powerful pattern recognition system should have powerful feature extraction and classification. Note that, feature extraction in gray level or RGB color space has problems such as environmental light changes, adding noise or changes in contrast and sharpness of images, which lead to weak classification. So the HSV color space was used. Here, Block-based Improved Center Symmetric Local Binary Pattern is introduced for feature extraction. In each component of the HSV color space, information of highlight areas in the image such as edge, shape and some texture was extracted. The histogram was calculated in two-level blocks and Support Vector Machine was used for classifying into vehicles, motorcycles and pedestrians. The obtained results in increasing the detection accuracy and decreasing the spent time were satisfactory.