A Constructive Review on Pedestrian Action Detection, Recognition and Prediction

Abstract

Analysis of pedestrian activities in the video sequences is an intriguing domain that incorporates vast applications, such as autonomous driving systems, traffic control systems and interactions between people and computers. The primary focus of this research was on evaluating several strategies to analyse pedestrian activities effectively. The constructive comparison included three main steps, i.e. detection of the pedestrian, recognition of their actions and prediction about the activity of the pedestrian. Changes in activities of pedestrians, dynamic background, moving camera, view angle and processing time made it more challenging. Recent approaches were justified and compared based on precision accuracy, processing time and minimum resource allocation. The results were also compared by a series of state-of-the-art research datasets with provided significant observations in terms of greater accuracy which can lead to the construction of an extremely improvised system that would save pedestrian people from road accidents and assist autonomous driving systems. The purpose of this study is to discuss the current progress using different approaches.