Extended Gradient Local Ternary Pattern for Vehicle Detection

Abstract

In recent years, many vehicle detection algorithms have been proposed. However, a lot of challenges still remain. Local Binary Pattern (LBP) is one of the most popular texture descriptors which has shown its superiority in face recognition and pedestrian detection. But the original LBP pattern is sensitive to noise especially in flat region where gray levels change rarely. To solve this problem, Local Ternary Pattern (LTP) is proposed. Nevertheless, LBP and LTP are lack of gradient information. In this paper, after analysis and comparison, we propose a novel feature descriptor named Extended Gradient Local Ternary Pattern (EGLTP). The proposed descriptor, Extended Gradient Local Ternary Pattern (EGLTP), contains properties of other features, such as the original LTP being less sensitive to noise, Semantic Local Binary Patterns (S-LBP) having low complexity and good direction property, and HOG including lots of gradient information. Experiments showed that EGLTP feature is very discriminative and robust in comparison with other features.