OHON4D: Optimized Histogram of 4D Normals for Human Behaviour Recognition in Depth Sequences

Mourad Bouzegza, Ammar Belatreche, Ahmed Bouridane, Mohamed Elarbi-Boudihir

Research output: Contribution to journalArticlepeer-review


Understanding human behavior in video streams is one of the most active areas in computer vision research. Its purpose is to automatically detect, track and describe human activities in a sequence of image frames. The challenges that researchers have to face are numerous and complex so that building a faithful feature vector that describes and identifies the human behaviour remains a crucial aspect. This paper presents a geometry-based descriptor whose features are extracted from data acquired by depth sensors. It uses a heuristic approach to optimize the HON4D (Histogram of Oriented 4D Normals) descriptor proposed by O. Oreifej and Z. Liu. The latter used a histogram to describe the depth sequence by extracting the normal orientation of the surface distribution in the 4D space of time, depth, and spatial coordinates. The proposed approach in this paper, called OHON4D (Optimized Histogram of 4D Normals), enhances the HON4D method by considering only four projectors to represent a 4D normal instead of 120. We obtained a similar accuracy while saving approximately half of the computational time.
Original languageEnglish
Pages (from-to)1-25
Number of pages25
JournalInternational Journal of intelligent Engineering Informatics
Publication statusAccepted/In press - 29 Dec 2023

Cite this