Neural network training method and three-dimensional gesture posture estimation method

Abstract

The present invention discloses a neural network training method and a three-dimensional gesture posture estimation method. The neural network training method comprises the steps of (S1) collecting adata set including multiple gesture depth images by a depth camera, (S2) using the data set in the step (S1) to train a random forest learner, (S3) using the random forest learner to segment the multiple gesture depth images in the data set in the step (S1) to obtain gesture sub images, processing the gesture sub images to obtain processed images, and carrying out random sequence division on the processed images and the multiple gesture depth images in the set in the step (S1) to form a training set and a testing set, and (S4) training a convolution neural network by using the training set andthe testing set obtained in the (S3) to obtain a network model. According to the three-dimensional gesture posture estimation method, the network mode is used to estimate a three-dimensional gestureposture in a single depth image. The concrete positions and states of palms and fingers in a gesture can be accurately identified.

Description

technical field [0001] The invention relates to the fields of computer vision and deep learning, in particular to a neural network training method and a three-dimensional gesture estimation method. Background technique [0002] In recent years, with the rapid development of computer vision and deep learning, virtual reality and augmented reality technologies have gradually become popular, and still have immeasurable development prospects. As an important means of human-computer interaction, gesture recognition technology has been highly concerned by the field of computer vision. Due to the many joints of the human hand, complex shapes, high degrees of freedom and easy occlusion, the gesture position can be quickly and accurately recognized. Hand movements have always been a problem. [0003] Traditional gesture estimation methods can generally be divided into two types: sensor-based and image-based. Sensor-based gesture and posture estimation technology refers to fixing se...

AI Technical Summary

Problems solved by technology

Sensor-based gesture and posture estimation technology refers to fixing sensors such as accelerometers and angular velocity meters on specific parts of the palm and fingers of a person; obtaining the position and motion state information of a specific part of the human hand through a wearable sensor device, and then using kinematics The method solves the state of the palm and fingers of the human hand, so as to achieve the effect of gesture posture estimation; this method has great limitations in gesture detection due to the need to wear sensor equipment, and is affected by factors such as the accuracy of the sensor itself and the change of the wearing position. The detection error will be larger

Another image-based gesture estimation method usually uses edge detection, skin color detection and other methods based on edge or area detection on images containing human hands captured by an RGB camera. First, determine the approximate area of ​​​​the human hand in the image, and then Segment detailed information such as fingers and wrists through image segmentation and other means; because ordinary cameras capture pictures containing hands, they usually only reflect the plane information of the scene. There is also a large error

Images