Abstract: | According to the feature of the fingers' geometry being roughly parallel, this paper
proposes a new method based on curvature and paralleled-vector for fingertip detection. This
method first uses depth image information from complex background environment to segment
hand quickly, and then based on morphological operations and median filter for smoothing, edges
can be extracted. Subsequently, we can use curvature to locate the position of the virtual
fingertips. Finally the paralleled-vector feature is adopted to rule out error detection points. The
experimental results show that the method in different finger motions, different light intensities,
and complex environment can locate the position of the fingertips. Besides, the fingertips
detection rate can reach 98.64%. |