Gesture recognition method

Abstract

The invention belongs to the technical field of gesture recognition, and relates to a gesture recognition method. The method comprises the following steps: acquiring a depth image sequence in real time via Kinect; acquiring the positions of a palm center and an elbow center in real time via the Kinect, and extracting a hand contour according to the depth information of each frame of image; calculating a fingertip position and a finger root position according to the palm center position, the elbow center position and the hand contour, and extracting features; matching the features extracted in step 3 with gesture features in a template library, entering a classifier, selecting the most proximate gestures as recognized gestures according to the classification standard of the classifier, and saving the gesture recognized in each frame into a queue Q; analyzing the recognition results of the frame and the front four frames in the queue Q, and selecting the gestures of a largest quantity as the final recognition result. The method can quickly and accurately recognize the gestures of an operator by acquiring depth information using Kinect and in combination with a digital image analysis technology.

Description

technical field [0001] The invention belongs to the technical field of gesture recognition, and relates to a gesture recognition method. Background technique [0002] Gestures are currently one of the most natural and intuitive ways of human-computer interaction. The existing gesture recognition technology is mainly divided into two types. One is gesture recognition based on data gloves, which require users to wear special data gloves to track and mark the movement trajectory and timing information of human hands in spatial coordinates, and then pass signal Processing to achieve the purpose of gesture recognition, this method requires the purchase of expensive equipment and has low practicability. The other is gesture recognition based on computer vision, which mainly uses ordinary cameras to obtain image data including hands, and recognizes gestures through data processing and analysis. This method has poor real-time performance, and the detection effect depends on the amb...

AI Technical Summary

Problems solved by technology

[0004] (1) Algorithms based on color images and skin color segmentation are susceptible to factors such as lighting conditions, complex backgrounds, and skin color changes. In an environment with insufficient lighting or hand color differences (such as doctors wearing gloves during surgery) Under this condition, the recognition efficiency will be greatly reduced, which will affect the practicability

[0005] (2) The recognition algorithm is complex and the amount of calculation is large, which cannot meet the real-time requirements

[0006] (3) In the case of gesture rotation, translation and scaling, the recognition rate is not high

Images