An approach for infrared image pedestrian classification based on local directional pixel structure elements' descriptor

Abstract

Pedestrian classification is a major problem in infrared (IR) images due to lack of shape, low signal-to-noise ratio and complex background. And it find applications in agriculture, forestry, night vision monitoring system, intelligence system and defence system. In this paper, local directional pixel structure elements descriptor (LDPSED)-based pedestrian classification approach is proposed to overcome these problems. In addition, for segment the objects (pedestrian and non-pedestrian) from an IR image interest point detection approach is proposed. The proposed method consists of three steps segmentation, feature extraction and classification. Firstly, objects are segmented from the input image. Secondly, the feature extraction is carried out on the segmented objects. Finally, support vector machine (SVM) is implemented for classification of objects in IR image into pedestrian and non-pedestrian. To prove the effectiveness of the proposed approach, we have conducted experimental test on the standard OTCBVS-BENCH-thermal collection over the OSU thermal pedestrian database. In addition, the classification results of the proposed approach are compared with the existing approaches. The efficiency of the proposed approach is proven by high classification accuracy.