A Web-based semantic tagging and activity recognition system for species' accelerometry data

详细信息    查看全文

• 作者：Lianli Gao^a ; ^{l.gao1@uq.edu.au} ; Hamish A. Campbell^b ; ^{hamish.campbell@uq.edu.au} ; Owen R. Bidder^c ; ^{367097@swansea.ac.uk} ; Jane Hunter^a ; ^{j.hunter@uq.edu.au}
• 关键词：Semantic annotation ; Tri-axial accelerometer data ; Animal activity recognition ; Support vector machines ; Visualization
• 刊名：Ecological Informatics
• 出版年：2013
• 出版时间：January, 2013
• 年：2013
• 卷：13
• 期：Complete
• 页码：47-56
• 全文大小：1704 K

文摘

Increasingly, animal biologists are taking advantage of low cost micro-sensor technology, by deploying accelerometers to monitor the behavior and movement of a broad range of species. The result is an avalanche of complex tri-axial accelerometer data streams that capture observations and measurements of a wide range of animal body motion and posture parameters. Analysis of these parameters enables the identification of specific animal behaviors¡ªhowever the analysis process is immature with much of the activity identification steps undertaken manually and subjectively. Consequently, there is an urgent need for the development of new tools to streamline the management, analysis, indexing, querying and visualization of such data. In this paper, we present a Semantic Annotation and Activity Recognition (SAAR) system which supports storing, visualizing, annotating and automatic recognition of tri-axial accelerometer data streams by integrating semantic annotation and visualization services with Support Vector Machine (SVM) techniques. The interactive Web interface enables biologists to visualize and correlate 3D accelerometer data streams with associated video streams. It also enables domain experts to accurately annotate or tag segments of tri-axial accelerometer data streams, with standardized terms from an activity ontology. These annotated data streams can then be used to dynamically train a hierarchical SVM activity classification model, which can be applied to new accelerometer data streams to automatically recognize specific activities. This paper describes the design, implementation and functional details of the SAAR system and the results of the evaluation experiments that assess the performance, usability and efficiency of the system. The evaluation results indicate that the SAAR system enables ecologists with little knowledge of machine learning techniques to collaboratively build classification models with high levels of accuracy, sensitivity, precision and specificity.