International Journal of Multimedia and Ubiquitous Engineering Vol.11, No.10 (2016), pp.175-182 http://dx.doi.org/10.14257/ijmue.2016.11.10.17 ISSN: 1975-0080 IJMUE Copyright ⓒ 2016 SERSC Pedestrian Detection Algorithm Combining HOG and SLBP Aili Wang 1 , Mingxiao Wang 1 , Jitao Zhang 1 , Yuji Iwahori 2 , and Bo Wang 1 1 Higher Education Key Lab for Measuring & Control Technology and Instrumentations of Heilongjiang, Harbin University of Science and Technology, Harbin, China 2 Dept. of Computer Science, Chubu University, Japan aili925@hrbust.edu.cn Abstract In order to solve the problem of pedestrian detection performance, the described operator was improved. In this paper, semantic local binary pattern (SLBP) and histogram of oriented gradient (HOG) are combined as new feature operator. This feature method would enrich the information and enhance the detection performance. And then histogram intersection kernel support vector machine (HIKSVM) classifier is trained by the augment feature. Because the time cost is too large by the conventional SVM. HIKSVM could make up this drawback, and significantly reduce the training time. The experiments on the INRIA pedestrian dataset show that the method obtained significant improvement in accuracy comparing to HOG descriptors. Keywords: Pedestrian detection; HOG; SLBP; HIKSVM 1. Introduction Pedestrian detection has very important applications in video surveillance, content- based image retrieval, video annotation and so on. However, detecting humans in images is a challenging task owing to their variable appearance and the wide range of poses that they can adopt. Pedestrian detection is to segment the pedestrian outline from the background and locate accurately in each frame of the video sequence. It is a typical challenging task in object detection field. However, owing to high variations of pose, clothing, cluttered backgrounds and partial occlusion handling that people can adopt, the detection task is rather difficult. So it is crucial to extract feature and choose classifier. Several prevalent features are widely used for pedestrian detection, such as Haar-like feature, Local Binary Pattern(LBP) , Histogram of Oriented Gradient(HOG) and Scale Invariant Feature Transform(SIFT), etc. Among these features, HOG is better than other features in pedestrian detection, and usually used to capture the edge or local shape information. The most representative work can be found in [1-5], where overlapped and dense local descriptors based on HOG are extracted trained via Support Vector Machine (SVM), and detected by classifying the images window. It gives significantly higher accuracy on INRIA human database. Moreover, they find that HOG combined with SVM is a better method in a compromise between the runtime and accuracy through great experiments. T. Ojala et al. [6] developed Local Binary Pattern(LBP) operator to extract local texture features, that is highly discriminative and its key advantages is namely invariance to rotation [9] and monotonic gray level changes [7]. Wang et al. [8] combined HOG and LBP to solve the partial occlusion problem. For the histogram is built according to binary-to-decimal conversion codes, there is no guarantee that semantically similar features must fall into spatially nearby histogram bins, so we adopt semantic LBP(SLBP)to replace basic LBP and combine with HOG feature as the method in this paper.