Recognition of human actions

Human action recognition aims to recognize the actions performed by humans in daily life, as the computer has high performance on computing and analysis, it helps in a lot of areas, such as security surveillance, interactive human-machine communication, etc. One of the most robust and fast recognition approaches is the Motion History Image (MHI) based temporal-template matching method. This project focuses on the usage of Motion History Image (MHI) and Motion Energy Image (MEI), traditional Hu Moments matching approach was tried to verify its recognition ability. Based on it, modification was implemented; instead of using whole MHI image to get Hu Moments, the MHI image of human body is segmented into 4 body parts – head, arm, torso and leg. Hu Moments approach was carried out for each body part and recognition was realized by combining effect of the 4 body parts. Although the modification improved the recognition performance, however, the increasing computational complexity and some subjective operations were not desired for the recognition. Finally based on MHI images, Local Binary Pattern (LBP) approach, which initially used in face recognition, was proposed and adopted to solve the problem in computational complexity and improve recognition performance.Bachelor of Engineerin