Research On Human Activity Recognition Ways Based On Wearable Sensors

With the development of the electronic and communication technology, wireless sensor has been widely applied. Human activity recognition based on micro inertial sensor as a new branch of artificial intelligence, has drawn much attention from hu-man beings. And it becomes more and more important. Compared with human activity recognition based on computer vision, it has advantages such as low complexity of equipment, being little influenced by external changeable environment and better spatial freedom. As an important application of Internet of things in the health care, health recovery and helping the elder and disabled, human activity recognition using wearable sensors has wide application prospects and very considerable economic benefits. The research in this field is a very meaningful work. However, there are some deficiencies in the current work of activity recognition based on sensors at home and abroad. This thesis is focused on features and classifier methods of activity rec-ognition, the main research works are as follows:Firstly, in consideration that machine learning and pattern recognition methods are the main methods used for activity recognition with sensors, theory of sparse re-presentation and compressive sensing are introduced to resolve the classification of activities. And in view of the multi-sensor activity recognition problems, an effective result fusion method can be proposed. In multi-sensor activity recognition system, in one way, a parallel processing architecture is helpful to improve the speed of activity recognition, and in another way, multi-result fusion method play a vital role in keep-ing or promoting behavior recognition rate. By building a residual error model, full information of activity is extracted. And this model can make a good performance of human activity recognition using sensors.Secondly, in consideration that features used by activity recognition based on sensors are time and frequency domain features, and time and frequency domain fea-tures are always used in digital information processing, a set of correlation feature is proposed. In multi-sensor activity recognition system, according to the characters of human activity, data of sensor nodes at different location can give different informa-tion of human activities. Correlation feature is a way to combine different location’s data of sensor nodes. It can provide more useful information about activities. And correlation feature can extract potential information of activities, which can be very useful for improving the accuracy of human activity recognition. Finally, with most researches in different conditions, two public activity recog-nition databases are used in experiments. It’s helpful to building a single uniform. And with a large number of simulation experiments, results show that the proposed methods can improve the performance of activity recognition effectively.