Smoking Gesture Recognition Using Visual Analysis Techniques

Abstract

Smoking gesture recognition in video-surveillance systems is still an open challenge for computer vision and pattern recognition communities because of its shape, motion, its dependency on lightning conditions, background disturbance and colors of the scene. Smoking gesture detection by visual clues can allow fast and reactive alarm systems as compared to smoke detection sensors. The proposed project is intended to provide similar services especially for the university environment. With the development of intelligent video surveillance systems, a relatively newer technology of gesture recognition has attracted considerable attention. The basic idea is to use image frames of a scene captured by a CCTV camera or any camera to determine if a person is performing a peculiar gesture which in this case is smoking. Various image processing and machine learning algorithms are applied to process the captured image frames in order to determine the smoking gesture within the vision field of the cameras. Distinguishing features like color, motion and distance are used to detect smoking gesture and determine whether a person is smoking or not. The proposed system can be easily incorporated into building video surveillance systems.