View-Invariant Hand Gesture Planar Trajectory Recognition on Monocular Vision

On monocular vision, in the process of the hand gesture recognition, when the camera poses are different, the same motion trajectory can project into different trajectory projections, which will affect the recognition and application of the trajectory. To solve this problem, using the square calibration, a plane projection’s standardization model is built and used in a new active vision based view-invariant gesture trajectory recognition method. Firstly, the square calibration feature points on the projected plane are extracted, and matched to the points on the ideal image plane. Secondly, using the proposed standardization model, the Homography between the two planes is obtained. Finally, from the re-projection, the view-invariant trajectory projection is obtained and recognized. Experimental results show that the standardized trajectory is similar to the orthographic projected trajectory. The recognition efficiency is improved by 20%. The processing speed in MATLAB environment is 48.86 frames per second, which meets the real-time requirement of the hand gesture interaction.