Comparison of HOG, LBP and Haar-Like Features for On-Road Vehicle Detection

Autonomous vehicles may be the most significant innovation in transportation since automobiles were first invented. Environmental perception plays a pivotal role in the development of self-driving vehicles which need to navigate in a complex environment of static and dynamic objects. It is required to extract dynamic objects like vehicles and pedestrians more precisely and robustly to estimate the current position, motion and predict its future position. In this article, the performance of three commonly used object detection approaches, Histogram of Oriented Gradients (HOG), Haar-like features and Local Binary Pattern (LBP) is investigated and analyzed using a public dataset of camera images. The detection results show that for the same dataset, LBP features perform better than the other two feature types with a higher detection rate. Finally, a unique and robust detection algorithm using a combination of all the three different feature descriptors and AdaBoost cascade classification is proposed.