Positioning Method of Pedestrian Dead Reckoning Based on Human Activity Recognition Assistance

Aiming at the problem that the traditional Pedestrian Dead Reckoning (PDR) cannot adapt to the reliable positioning of the target in different motion states, this paper designs and proposes a positioning method based on deep learning for Human Activity Recognition (HAR) assisted PDR. First, the Wavelet-CNN deep learning network is used in the offline stage to preprocess and train the data of the built-in MEMS sensor of the smartphone to obtain the HAR model. Then, in the online real-time positioning stage, the different motion states of the target are identified based on the HAR model, and the pedestrian step detection and step size estimation algorithms are adaptively adjusted. Finally, the HAR-assisted PDR algorithm is implemented on the smart-phone, and a large number of tests and verifications are carried out in the experimental environment. The proposed localization method for HAR-assisted PDR based on deep learning can accurately identify a variety of complex human motion states, and the recognition accuracy is as high as 99.50%. At the same time, the accuracy rate of the step recognition algorithm of state-of-the-art PDR is increased by 10.94 %, and the maximum positioning error is reduced by about 16.2%, which verifies the effectiveness of the proposed algorithm.