Human activity recognition based on fusing inertial sensors with an optical receiver

The research on Human Activity Recognition (HAR) systems has received high attention due to its importance in high demanding and challenging fields of study such as health care, social science, robotics and artificial intelligence. One of the most prominent approaches is to use Inertial Measurement Unit (IMU) sensors in order to determine what activity a human is making. If complex activities such as sit-down, stand-up, walk-up and walk-down are needed to be recognized, the user needs to wear multiple sensors on his/her body to perform a correct recognition. Such activity recognition will be of high interest if the object's position is also recognized. For recognizing the activity and location properly, a decent fusion technique between the multiple sources of information is required. In this study, we propose a novel positioning and HAR system based on fusing data from a single IMU device with data from a simulated segmented optical receiver to perform visible light positioning (VLP). We combine real world data collected from the IMU device with optical simulation data generated from a simulated segmented optical receiver in order to distinguish between various complex activities, particularly walk, walk-up and walk-down in addition to determining the position of where the activity is performed. The fusion mechanism does not only improve the accuracy of the activity recognition in comparison to utilizing either IMU or optical data alone, but also enables the system to furthermore retrieve the user's position in the room. By applying different Machine-learning (ML) algorithms for the assessment of the achievable results, we conduct a comprehensive analysis on which ML method is suitable for our envisioned low-complex HAR and positioning system, which avoids the placement of multiple sensors on the user's body. Our results show the influence of different segmentation strategies for the novel concept of a segmented optical receiver in combination with an IMU sensor on the accuracy of the activity and position recognition.