Pedestrian Attributes and Activity Recognition Using Deep Learning: A Comprehensive Survey

Abstract

In recent years, pedestrian attributes and activity recognition has attracted expanding research emphasis due to their considerable research importance and value of application in the intelligent civil and military domains. Owing to the inadequate image or frame quality of inexpensive cameras and the absence of obvious and stable feature information, direction of pedestrian movement, and so on, the complication of pedestrian attributes and activity recognition is expanded. However, with the comprehensive implementation of deep learning techniques, pedestrian attributes and activity recognition has made a substantial advance. This paper is the first of its kind because it combines the recognition of pedestrian attributes and activities separately, and reviews the works on pedestrian attributes and activities using deep learning in relation to datasets. The fundamental concepts, corresponding challenges, and popular solutions are also explained. Furthermore, in this community, metrics of evaluation and concise performance comparisons are given. In the end, the hotspots of the present research and the directions of the future research are summarized.