Visual-image Based Finger Joint Angle Measurement Research And Implementation

Visual-image based gesture recognition is a hotspot of artificial intelligent research. Traditional methods of gesture recognition are incompetent to detect finger’s delicate movement due to poor segmentation effect using2D images and limited gesture templates through supervised training of classifier.In recent years, the development of hardware and the implementation of3d camera technology make detection of complex gesture in high precision become possible and turn it from rough estimation into precise measurement. This paper proposed a method of finger joint angle measurement based on visual images. The main work in this paper is as follows:1. Study Kinect device HW/SW to get reliable3d images input for finger joint angle measurement.2. Study skin detection algorithm based on region growing, optimizes the basic algorithm by integrating it with depth information to make detection more adaptable for complex background and environment.3. Propose hand modeling to overcome high dimension problem, reduce the DOF number to estimate without accuracy loss.4. Study OpenGL3D rendering technology, propose many new and creative methods to accelerate rendering.5. Study OpenCL parallel computing technology and GPU HW, design proper objective function for pixel-wise calculation.6. Optimize PSO to calculate the best model and analyze the measurements.