Hand gesture recognition in low-intensity environment using depth images

In this paper, a method is proposed for the recognition of hand gesture using depth image postures. It supports a vision-based gesture recognition driven system capable of recognition in low-intensity environments. The algorithm processes video in real-time and generates instantaneous output in the system. Firstly, we obtain depth images, and generate disparity map and perform masking. Secondly, nearest contour extraction is performed followed by polygon approximation. Further, convexity defects for counting fingers are obtained. Aspect ratio (height-to-width ratio) analysis is employed to refine accuracy of results. The number of fingers raised is displayed and stored in a variable. For different values of the variable: 0, 1, 2, 3, 4, 5, different appliances can be controlled. A single value can serve as a toggle switch for an appliance. We have ensured robustness of our program by testing it in various backgrounds and lighting conditions. The program has been scripted in Python, using the OpenCV and Open Kinect Libraries and synchronized through Linux with Kinect. When lights are ON, both RGB and IR cameras are in play. When lights are OFF, only IR bit stream is active. Depth-based segmentation makes precise identification of the region of interest. We have achieved satisfactory accuracy levels of above 90% with real-time processing. We have optimized the algorithm for appropriate distances. In the future, we intend to expand our algorithm for a variety of gesture control applications in software and hardware and take advantage of its blazing fast computation and real-time processing.