Pedestrian Detection Techniques: A Concise Review

Pedestrian detection is a sub-branch of object detection with multiple applications, notably in autonomous driving and assisted driving systems. However, it faces many challenges, such as occlusion, scalability, and complex backgrounds. This paper aims to summarize recent state-of-the-art progress in pedestrian detection technology. First, it gives a general understanding of a pedestrian detection system and then delves into ethical considerations such as pedestrian privacy. It then categorizes pedestrian detection techniques into three main categories: classical methods, deep learning methods, and hybrid methods. Deep learning methods are further subdivided into three types: anchor-based, which uses predefined anchor boxes to detect pedestrians and includes both one-stage and two-stage detection; anchor-free; and attention-based methods. Secondly, it provides a literature review of various papers on pedestrian detection.