Gait recognition using kinect and locally linear embedding

Abstract

This paper presents the use of locally linear embedding (LLE) as feature extraction technique for classifying a person’s identity based on their walking gait patterns. Skeleton data acquired from Microsoft Kinect camera were used as an input for (1). Multilayer Perceptron (MLP) and (2). LLE with MLP. The MLP classification accuracy result was used for comparison between both. Several MLP and LLE properties were tested to find the optimal number of setting that can improve the MLP performance. Based on the two methods used, the neural network implemented with LLE showed the better accuracy compared to the neural network alone.

Keywords: locally linear embedding; neural network; multilayer perceptron.