Human Motion Posture Recognition Base On Accelerometer

In recent years, accelerometer based human motion posture recognition is an emerging and active research in the field of computer pattern recognition. More and more researchers monitor human motion posture by using accelerometer in laboratory and natural environment. Human motion posture contains accelerometer data acquisition, data processing, features extraction and selection, classifier design and so on. Some researchers prefer to focus on one part of research or the other. A lot of research findings have received after constant efforts of researchers. This research has a lot of very important issues and difficult problems to conduct research, such as, quality data acquisition, the extraction method of effective features, and the high efficient classifier design. The research of feature extraction and classifier design has done in this paper is listed as follows:1. Because the research on human motion posture recognition is still in the primary stage, there is still no standard dataset which meet all research requirements of recognition system. So, we design a simple data acquisition system. A data collector with one tri-axial accelerometer located on the subject’s waist, and each subject asked to perform six motion postures in laboratory. According to different activity has different change speed of acceleration, two features (front and rear subtract, and approximation slope) have been proposed.2. A honey-bee mating optimization random forest algorithm via honey-bee mating optimization was proposed. First, we extract five features of acceleration signal. Then, lots of random forests (namely bee) are built, and a best queen bee is selected to create an optimal classification model. Finally, the improved random forest algorithm is used to classify the motion postures. We investigate the using frequency of all features, and compare the performance of random forest, support vector machine and our algorithm in the same experiment. The experimental results have confirmed the proposed algorithm better than the other algorithms, but some activities are confused because of noise data.3. A human motion posture recognition method integrates with curve fitting and k nearest neighbor is presented. First, the acceleration data is divided into isometric groups with the same activity. Then, curve fitting in least-square method is used to change the data into curves. Finally, curve similarity is used to calculate the distance between curves, and k nearest neighbor is adopted as a classifier to recognize human motion posture. The experimental results have confirmed that the proposed algorithm is feasible and effective.Overall, accelerometer based human motion posture recognition is still in the developmental stage. This research subject is of important theoretical value and increased application requirement. So, it is worth carrying out more effective and further research, and may be applied in practice.