A method for close-range gesture recognition based on morphology

Abstract

The present invention aims to propose a method for recognizing close-range gestures based on morphology. First, collect RGB color images and depth images, and use the haar feature detector to obtain the hand mask box1 in the color image; then obtain the RGBD image of the segmented hand area according to the RGBD alignment principle, first-order difference and threshold processing, and crop them. Lower the region of interest ROI to obtain the hand mask box3; then solve the maximum inscribed circle for the hand mask box3, and estimate the palm size and palm position through the geometric parameters of the inscribed circle; design a coordinate elimination mechanism to extract the fingertip area and the number of fingertips N, and select the pre-trained CNN classification model according to the value of N, classify the gestures, and obtain the final gesture type. In this method, a coordinate elimination mechanism and a new scientific image convolution operator are specially designed in the process of recognizing the hand region. The operator has rotation invariance and can quickly recognize complex gestures in close range.

Description

technical field [0001] The present invention relates to algorithms for gesture recognition. More specifically, it is a method of close-range gesture recognition based on morphology, and a specific method to realize this method to accelerate the calculation is proposed. Background technique [0002] With the wide application of computers, human-computer interaction has become an important part of people's daily life. The ultimate goal of human-computer interaction is to realize natural communication between humans and machines, so gesture recognition research conforms to the needs of people's life development. At present, gesture recognition has also been applied in various fields, such as live video, robotics and AR. However, due to the diversity, ambiguity, and differences in time and space of gestures, as well as the complexity of the human hand and the ill-posedness of vision itself, gesture recognition has become a multidisciplinary research topic. [0003] There are ...

AI Technical Summary

Problems solved by technology

This method involves an effective and reasonable coordinate elimination mechanism in the process of recognizing the hand area, including a new type of scientific image convolution operator, which is invariant to rotation and can quickly recognize complex gestures in the near view. Strong filtering ability, which solves the difficulty of identifying and merging fingers with existing algorithms

Images