Approach to hand posture recognition based on hand shape features for human–robot interaction

Abstract Hand segmentation is the initial step for hand posture recognition. To reduce effect of variable illumination in step, a new CbCr-I component Gaussian mixture model (GMM) proposed to detect skin region. The region selected as interest from image using detection technique based on presented GMM and adaptive threshold. A shape distribution feature described polar coordinates extract contour features solve false recognition problem some shape-based methods effectively recognize cases when different postures have same number outstretched fingers. multiclass support vector machine classifier utilized posture. Experiments were carried out our data set verify feasibility method. results showed effectiveness approach compared with other methods.