PCA/ICA-based SVM for fall recognition using MEMS motion sensing data

This paper presents the progress towards a fall recognition algorithm based on MEMS motion sensing data. A Micro Inertial Measurement Unit (muIMU) that is 66 mm times 20 mm times 20 mm in size is built. This unit consists of three dimensional MEMS accelerometers, gyroscopes, and a Bluetooth module. It records human motion information, and the database of FALL and NORMAL is formed. We propose principal component analysis (PCA) for feature generation and independent component analysis (ICA) for feature extraction. Then, we use support vector machine (SVM) for training process. Experiments show that the process can classify falls and other normal motions successfully.