Feature Based Human Posture Recognition System
Loading...
Date
2018
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Management & Technology
Abstract
With the increase in the population of elderly, especially in the developing countries, the
increase in the number of aged people living independently has arose a great risk. It has led to
increased health care financial cost which can become a great burden on our society. The falls
of elderly has become a major cause for serious injuries. Due to this risk many studies have
been done on human posture recognition, its different approaches and how we can predict the
fall before any serious injury occurs to the elderly people. This work comprises of three parts.
Firstly, this work provides an extensive survey on human posture recognition systems with
respect to eras, different approaches used in previous researches and performance evaluation
of different datasets used for human posture recognition. Secondly, in this work a new
approach for human posture recognition is proposed which uses Radon Transform as a
feature extractor to extract projection feature in an image frame. The feature extracted using
radon transform will be used to train the machine learning classifier for postures
classification. Thirdly, classifier boosting is performed on the machine learning classifiers to
increase the overall accuracy of the proposed approach. Total five postures are recognized
using this approach which are walking, sitting, lying on the sofa, lying on the ground and
crouching. This thesis also provides the performance comparison between a recent existing
approach and the proposed approach. The proposed algorithm has achieved an accuracy of
98.7% on MCF (multiple cameras fall) dataset.
Description
Dr. Malik Tahir Hassan
Keywords
Gaussian Mixture Model, Karmen Loeve, Markov Random Field, MS