Human body posture estimation method and recognition device

Abstract

The invention provides a human body posture estimation method and a recognition device. The method comprises the steps: a preprocessed image is transmitted into a high-resolution network, and a feature map isgenerated at each branch of the network through down-sampling, up-sampling and convolution operation; and an attention module is connected behind the generated feature map to capture key features, a new feature map is generated by fusing the feature map at the final stage with the feature map at the previous level, the new feature map is recovered to the resolution which is the same as that of a trunk through deconvolution operation, a final heatmap is obtained, and the position of each joint point can be predicted according to the heatmap. The image is processed by the high-resolution network, so that effective information loss can be avoided; feature information can be efficiently utilized by adding the attention module; the problems of inaccurate occlusion image attitude estimation and difficult small target prediction can be solved by interaction information of multiple resolution branches; and real-time detection visualization can be realized by transmitting an output result into the device.

Description

technical field [0001] The present application relates to the technical field of human-computer interaction, and in particular to a method for estimating human body posture and a recognition device. Background technique [0002] Human body pose estimation technology refers to the labeling of the joint positions of the human body in ordinary images or videos. According to the number of labeled objects, human pose estimation can be divided into single-person pose estimation and multi-person pose estimation. The single-person pose estimation task can be regarded as the joint position detection for determining the individual, and the joint position positioning is performed after finding the portrait position in the image. ; Multi-person pose estimation is more complex than single-person pose estimation. The existing technology mainly has two methods: top-down and bottom-up. Among them, the top-down method is to detect the portraits in the image first, and then perform single-pe...

AI Technical Summary

Problems solved by technology

[0005] This application provides a human body pose estimation method and recognition device, which can solve the problems of loss of effective information, inaccurate recognition of occluded images, poor real-time performance of pose estimation algorithms, and low degree of visualization of recognition results in the process of resolution restoration.

Images