Classification of Human Posture and Movement Using Accelerometer Data

Bodor et al. developed a novel method for employing image-based rendering to extend the capability of the human movement classification [1]. Sminchisescu et al. have developed algorithms for recognizing human motion in monocular video sequences, based on discriminative conditional random fields (CRFs) and maximum entropy Markov models (MEMMs) [2]. However, vision-based systems have issues in camera installation, lighting, picture quality, privacy and etc., which may render the system impractical in certain applications. As a branch of movement classification techniques, wearable sensor-based system could solve the above problems thanks to the development of nanomanufacturing technologies and ultralow-power embedded systems, which makes the wearable sensors cheap, small and compact. Ugulino et al. collected human movement data from 4 ADXL335 accelerometers, utilized C4.5 tree, Iterative Dichotomiser 3 (ID3) and AdaBoost to classify the human movements and have obtained high recall and precision [3]. This project utilizes the same set of data but a different set of models to compare classification performance.