Classification for the Dangerous and Non-Dangerous Physical Actions with SVM

• Main Authors: Ding-Jia Chung, 鍾定家
• Other Authors: Shing-Hong Liu
• Format: Others
• Language: zh-TW
• Published: 2011
• Online Access: http://ndltd.ncl.edu.tw/handle/56782043124729023073

碩士 === 朝陽科技大學 === 資訊工程系碩士班 === 99 === As advances of medical technology, the average age of people is growing. The relative age structure of population is ageing. Fall accidents mostly happen at the group of above 65-year elders which has 50 percent people who have occurred the fall accident and injured. Now, the three-axis accelerometer has been widely used to monitor the fall actions. In this study, we would classify the dangerous and non-dangerous physical actions which all belong to the body gravity going down. In training phase, the subjects have the 10 normal young people, including five men and five women. In testing phase, the extra subjects also have the 10 normal young people, including five men and five women, and 5 elders whose age is above 70 years. Using the intermittent actions and continuous actions test the performance of our method. The three-axis accelerometer was placed on the waist. The recorded signals were transforms to four parameters as the input vector. Support Vector Machines (SVM) was used to classify the dangerous and non-dangerous physical actions. We get the results of four parameters having the best performance, 98.4% of accuracy, 97.4% of sensitivity and 99.3% of specificity for the intermittent actions. In continuous actions test, the same training sample got false positive rate of 0.007 times / min and the elderly got false positive rate of 0.003 times / min. Moreover. For the another testing sample, 10 young people, they do five minute continuous actions including three times dangerous. The results show the accuracy rate being 80% to detect the dangerous actions.