Abstract:

Compared with the local binary pattern (LBP), the Haarlike LBP (HLBP) can effectively reduce the noise by using local statistics in pedestrian detection. However, when calculating the feature value in the HLBP texture, since the center point is not involved, its information is lost. In order to solve this problem, we propose an improved HLBP (IHLBP) texture feature method, which includes the central point in calculating the feature value with a maximum weight. We realize the two level decomposition operation via the twodimensional discrete Haar wavelet transform before the IHLBP feature extraction, and obtain three images with different scales. The IHLBP features of the three images of different scales are then extracted and normalized. They are series connected and finally the final feature vectors are obtained. Experiments on the INRIA Person data set using support vector machine (SVM) show that the proposed method can effectively improve the recognition rate of pedestrian detection.

Key words: pedestrian detection;image texture;IHLBP features;twodimensional discrete Haar wavelet;support vector machine