Framework for Traffic Personnel Gesture Recognition

Abstract

Human hand gestures can be used as most essential communication tool for human computer interaction (HCI). This paper presents a novel and efficient framework for traffic personnel gesture recognition based on Cumulative Block Intensity Vector (CBIV) of n-frame cumulative difference. The experiment carried out on the real time traffic personnel action dataset using Support Vector Machine (SVM), Decision Tree (J48) and Random Forests (RF). Experimental results denote the higher performance 96.24% of the Random Forests classification, when compared to the SVM and Decision Tree using 5-frame cumulative difference with CBIV features. The main part of this paper is the application of incremental SVM and tree based classifier techniques to the problem of identification of traffic personnel hand signals in video surveillance, based only on person hand movement.