Energy Saving in Forward Fall Detection using Mobile Accelerometer

State Of The Art

The current fall detection methods can be basically categorized as following:

• •

  Video data: The video-based system captures the images of human movement. A fall event is detected based variations of images (Toreyin, 2006);

• •

  Acoustic data: Detecting a fall via audio signals analysis;

• •

  Wearable sensor data: Embedding some micro sensors into clothes, or girdle, shoes, plugged on foot, etc. to monitor the human activities in real-time, and finding the occurrence of a fall based on the changes of some movement parameters (Winters, 2002).

However, these approaches have some weak points. For example, the narrow scope of video system, notably inaccurate of acoustic data, the inconvenience of wearable sensors system, and the limited range of application where they can be deployed at this moment. Accelerometer has been basically used in research (Wiliams, 1998). Since then, many studies have been presented about various methods using heuristic classifier (Bussman, 2001), while others used multiple sensors such as accelerometer and sound data (Charalampos, 2008). The common and simple methodology for fall detection is using threshold algorithm based on decision whether an event crosses predefined thresholds (AlertOne, 2008). Fall detector is also implemented on waist-worn accelerometer fixed on human body (Zhang, 2006) using SVM classifier. Mobile devices have some limitations on computational resources unlike full-fledged processing devices with wearable sensors as detection engine. Directly adopting wearable sensors research to mobile phone is not suitable as they are different in daily usage context (Korpipää, 2008). Proposing a fall detection algorithm that enables application to be lightweight for guaranteeing mobile phone resource’s consumption and also maintaining excellent accuracy is considered as a necessity at the moment.