A monocular camera based person-specific fall detection system exploiting deep neural network aided unsupervised

Yu, Miao and Gong, Liyun and Clifford, Ross and Duff, Carol and Ye, Xujiong and Kollias, Stefanos (2019) A monocular camera based person-specific fall detection system exploiting deep neural network aided unsupervised. In: IEEE International Conference on Acoustics, Speech, and Signal Processing.

Documents
A monocular camera based person-specific fall detection system exploiting deep neural network aided unsupervised

Request a copy
[img] PDF
icassp2019.pdf - Whole Document
Restricted to Repository staff only

611kB
Item Type:Conference or Workshop contribution (Paper)
Item Status:Live Archive

Abstract

In this paper, we propose a novel fall detection system based on a monocular camera. A single camera covering the living area is used for video recordings of an elderly person’s daily activities. From the video recordings, the human silhouette regions in every frame are extracted from the background, with related low-dimensionality representative features being automatically extracted by a convolutional neural network based autoencoder (CNN-AE). Extracted features together with the silhouette position information are combined together to construct an one class support vector machine (OCSVM) model for fall detection. From the comprehensive experimental evaluations in a real home environment,it is shown that the proposed fall detection method can detect different types of falls (falls in different orientations, falls with occlusions in a home environment) with high accuracy while achieving better performance than hand-crafted features based ones.

Keywords:fall detection, unsupervised learning, convolutional neural network autoencoder, one class support vector machine
Subjects:G Mathematical and Computer Sciences > G760 Machine Learning
G Mathematical and Computer Sciences > G400 Computer Science
Divisions:College of Science > School of Computer Science
ID Code:34275
Deposited On:27 Nov 2018 08:47

Repository Staff Only: item control page