Fall Detection System Using Kinect And Smartphone

With the continuous growing of the population of the older adults and the proportion of small family structure, falls, as the major killer of the elder, become more and more severe. Elderly people aged 60 and above has already reached 160 million in china by 2014, and by 2050, the proportion of elderly population will account for a third. Due to the ratio of living-alone or empty-nest families rises constantly, it become increasingly more difficult to rescue the elderly when there is an accident. About the elderly fall accident, for saving the life and health of the elderly it is very important to rescue the elder when there is an accident. The state-of-the-art fall detection systems are mainly based on the wearable devices and ambient and camera-based. The detection system based on wearable devices has a high accuracy but requiring people to wear the sensors day and night. Ambient fall detection systems need complex equipment and exorbitant price, and easily influenced by the outer environment. The camera-based fall detection systems have high accuracy in monitoring, but it is sensitive to environment light and it is not conducive to protect the privacy. So a low-cost automatic system capable of detecting fall accidents indoor is critical for older adults living.This dissertation, developed a fall detection system aimed to provide service for the elderly, this system can distinguish falls from activities of daily living accurately and inform the guardian at the first time of the fall accident. This dissertation, utilizing Kinect (a low-cost depth camera) and popular smart phone to obtain the motion information of the user, proposes a HON4D+ feature extracting method to analyze the depth-image sequence captured by Kinect in real time to achieve the human-action feature. Then, the human-action feature merged with the waveform sequence of the tri-axial accelerometer on the smart phone will be evaluated by a well-trained SVM classifier to decide whether the fall accident happens or not. The guardian will receive a phone call and SMS message including the accident video to launch an emergency measure at the first time of the fall accident from local area network. Through the experiment of sensitivity and specificity, experimental results show that the proposed system can achieve the fall accident detection accuracy up to 94.05%.