Foot Parameters Extraction using Deep Learning based Regression Model

The online shopping market is growing day by day due to advantages such as convenience of purchase and product accessibility. However, in the case of shoes, it is difficult to select shoes that can provide a comfortable fit by reflecting the characteristics of different feet for each user without direct wearing. In order to overcome this problem, various studies for extracting foot parameters are being conducted, but they have to use expensive equipment or are staying in the stage of extracting simple parameters such as foot length and width. In this paper, we propose a method to extract 20 foot parameters using a commercial depth camera-based scanning stage and deep learning. A scanning stage with 3 depth cameras captures the 3D point cloud of user's foot shape. The proposed method measures 6 foot parameters by analyzing the sole surface and the side surface of the point cloud and then estimates 14 foot parameters by using a deep regression model from 6 foot parameters. Validation experiments were carried out to assess the accuracy and feasibility of the system. The average error rate for foot parameter estimation is 4.527%.