Vision-based Air-writing Character Segmentation and Reproduction

To address the writing needs of visually impaired individuals, this paper proposes an air-writing segmentation and reproduction system that combines the advantages of air-writing with the efficiency of robotic learning by demonstration (LID). The solution utilizes the Mediapipe Hand to record the writer's continuous spatial character trajectory in front of the camera, and later uses a rule-based approach to achieve automatic segmentation of the continuous character trajectory. In addition, the utilization of dynamic movement primitives (DMPs) maintains the writer's original writing habits while reproducing the char-acters in a designated area, adhering to the writing specifications and ensuring legibility. The proposed system was tested on 15 sets of continuous word trajectories covering 26 letters, experimen-tally proves the effectiveness of the algorithm.