Gait recognition using principle component analysis implemented on DSP Processor

Abstract

This research focus on the development of an automatic human identification system using gait sequence images. Human identification is widely used in computer vision applications such as surveillance system, criminal investigations and human-computer interaction. Many identification approaches have shortcomings thus they require subject cooperation and sensitive to environmental and physiological changes. They also have high computational cost and are time consuming thus difficult to implement in hardware. Gait sequence consists of non-stationary data and can be modeled using a statistical learning technique. The proposed method consists of three different stages. The pre-processing stage computes the average silhouette images to capture the important information and get a better representation for gait silhouette data. Then a principle component analysis (PCA) technique is applied on the average silhouette to extract the important gait features and reduce a dimension of gait data. A linear projection method used in this stage is able to reduce redundant features and remove noise from the gait image. Furthermore, this approach will increase a discriminating power in the feature space when dealing with low frequency information. Low dimensional feature distribution in the feature space is assumed to be Gaussian, thus the Euclidean distance classifier can be used in the classification stage. The proposed algorithm is a model-free based which uses gait silhouette features for the compact gait image representation and a linear feature reduction technique to remove redundant information and noise. The proposed algorithm has been tested using a benchmark CASIA dataset. The experimental results show that the best recognition rate is 90% when the image is represented using 500 PCA coefficients. Low number of PCA coefficients will give a possibility for the Euclidean distance classifier to be implemented in hardware such as DSP processor. The implementation of the proposed algorithm using the DSP-based processor achieved better performance in term of computational time compared to the PC-Based processor with a ratio of 0.5 seconds.