A Deep Learning Approach to Automatically Extract 3D Hand Measurements

Accurate hand measurement data is of crucial importance in medical science, fashion industry, and augmented/virtual reality applications. Conventional methods extract the hand measurements manually using a measuring tape, thereby being very time-consuming and yielding unreliable measurements. In this paper, we propose-to the best of our knowledge-the first deep-learning-based method to automatically measure the hand in a non-contact manner from a single 3D hand scan. The proposed method employs a 3D hand scan, extracts the features, reconstructs the hand by making use of a 3D hand template, transfers the measurements defined on the template and extracts them from the reconstructed hand. In order to train, validate, and test the method, a novel large-scale synthetic hand dataset is generated. The results on both the unseen synthetic data and the unseen real scans captured by the Occipital structure sensor Mark I demonstrate that the proposed method outperforms the state-of-the-art method in most hand measurement types.

Automated summary

Accurate hand measurement data is crucial in medical science, fashion industry, and augmented/virtual reality applications. This paper introduces a deep-learning-based method to automatically measure the hand in a non-contact manner, outperforming existing methods in various hand measurement types.