Two-dimensional human skeleton point locating method for monocular depth video

Abstract

The invention relates to a two-dimensional human skeleton point locating method for a monocular depth video. The method comprises model building, a training process and an identification process. The training process comprises the following steps of S21, collecting a training sample; S22, generating a training target; S23, randomly initializing a deep model (a long-short-term memory convolutional neutral network) and model parameters, wherein the parameters consist of parameters of a convolutional layer and parameters in a long-short-term memory convolutional layer; and S24, by adopting a model optimization algorithm, updating the parameters of the deep model by utilizing the training sample in an end-to-end manner. The identification process comprises the following steps of S31, inputting monocular depth video frames; and S32, inferring and predicting human skeleton point positions by utilizing the trained deep model. By setting a reasonable learning target, the deep long-short-term convolutional neutral network is established; human skeleton point prediction features are obtained by adaptive learning according to a data driving mode; and a stable and reliable human skeleton point locating effect is achieved.

Description

technical field [0001] The invention relates to the fields of two-dimensional human posture recognition, computer vision, pattern recognition and human-computer interaction, and in particular to a method for locating two-dimensional human skeleton points in monocular depth video based on a long short-term memory convolutional neural network. Background technique [0002] Human skeleton point positioning is an important research direction in the field of computer vision research. Its main task is to enable the computer to automatically perceive the "what posture" of the person in the scene. It is widely used in home entertainment, action recognition, intelligent monitoring, patient care, etc. Monitoring and other systems that require human-computer interaction. [0003] The goal of human pose estimation is to automatically predict the pose data of each part of the human body (ie, the coordinates of the bone points in the image) from a two-dimensional image sequence. Due to t...

AI Technical Summary

Problems solved by technology

Due to the influence of environmental factors such as illumination and complex background in ordinary RGB color images or videos, it is very difficult to predict human skeleton points from monocular color images, and it is difficult to achieve robustness.

Images