A Preliminary Study on Personal Identification Using a 2D LiDAR and Machine Learning

Abstract

In this paper, we proposed person identification system based on gait pattern. As a preliminary study, classification was performed only on registered persons. The system uses the 2-dimention Light Imaging Detection and Ranging （LiDAR） to get gait feature. We use the LiDAR to measure the temporal change of distance between the LiDAR and body parts of a person walking towards below the LiDAR. The measured distance is drawn as a RGB color surface plot image showing the gait feature. Vertical and horizontal axes of the plot are time and angular direction viewed from the LiDAR respectively, and the distance is represented by color of the surface. We use deep learning approach to classify images. Convolutional Neural Network （CNN） is employed to predict person who passed below the LiDAR. We obtained 300 images per each of the 12 persons. Those 3,600 images were used for training of the CNN （2,400 images） and testing （1,200 images）. As results, our proposed identification system outperformed those 12 persons classified with accuracy of 0.89.