Activity Recognition Based On Triaxial Accelerometer

Activity recognition based on accelerometer is an emerging research direction in the field of activity recognition, it has the advantage of strong anti-interference ability, easy to carry, and free data acquisition method compared to vision-based activity recognition. This paper focuses on the acceleration sensor signal denoising methods, feature extraction and classification method has launched a series of studies, the main tasks include:1. A new threshold function is proposed for the disadvantage of hard-threshold and soft-threshold method in the wavelet denoising, it overcomes the oscillation of hard-threshold method and the fixed attenuation of soft-threshold method. Simulation shows that the proposed method can more effectively filter out the noise in the acceleration sensor signal compared to the hard-threshold and soft-threshold method.2. A recognition algorithm based on the standard deviation, skewness, kurtosis and correlation coefficients is proposed to recognize five walk patterns (standing, walking, running, upstairs and downstairs). The experiment shows that the proposed feature extraction method can effectively distinguish standing, walking, running, walking upstairs and walking downstairs, the average recognition rate is98%using support vector machine.3. A walking pattern recognition algorithm based on wavelet energy and interquartile is proposed to deal with the problem that it is hard to distinguish walking slowly, walking and walking quickly. The experiment shows that the proposed feature extraction method can achieve an average recognition rate of100%,9%higher than the traditional FFT coefficient, which shows the validity of the proposed feature extraction method.4. A triaxial accelerometer based gait recognition is preliminary studied, a gait recognition algorithm based on a variety of time-domain and time-frequency characteristics is proposed. The average recognition rate is95.7134%using support vector machine, indicating that a triaxial accelerometer based gait recognition is feasible and effective.5. For the disadvantages of traditional parameter selection method of support vector machine, we using particle swarm optimization to optimize support vector machine. The experiment shows the superiority of the particle swarm optimization based parameter selection method compared to the traditional parameter selection method.In all, accelerometer based activity recognition is an important research direction in the field of activity recognition. The research of this subject has great theoretical and practical significance, and it is worthy of detailed, in-depth research.