Single-person posture estimation method based on novel high-resolution network model

Abstract

The invention discloses a single-person posture estimation method based on a novel high-resolution network architecture. The method comprises the following steps: firstly, detecting an input image containing a single pedestrian by using a detector, removing an inaccurate detection box, and then expanding a data set through data enhancement; secondly, keeping a high-resolution feature map in an instantiated network structure through the parallel multi-resolution subnets without recovering the resolution, introducing exchange units into the parallel subnets, wherein each subnet repeatedly receives information from other parallel subnets, and the accuracy of single-person posture estimation is improved. In most complex scenes; the key points are shielded, so that a data enhancement scheme using one key point for shielding is provided, the trained convolutional neural network can be finely adjusted very effectively through the scheme, the shielded key points are positioned strongly throughadjacent matching, the accuracy of the shielding problem is improved, and a better model is obtained.

Description

technical field [0001] The invention relates to a method based on a novel high-resolution network architecture, which belongs to the interdisciplinary technical fields of deep learning, computer vision, and machine learning. Background technique [0002] 2D human pose estimation has been a fundamental but challenging problem in computer vision, and the goal of single-person pose estimation is to locate key points (e.g., elbow, wrist, etc.) or parts of human anatomy. The application background of attitude estimation is very extensive, mainly focusing on intelligent video surveillance, human-computer interaction, virtual reality, smart home and so on. This patent is interested in single-person pose estimation, which is the basis for other related problems, such as multi-person pose estimation, video pose estimation, and behavior recognition and tracking. [0003] Recent developments show that deep convolutional neural networks have achieved the highest accuracy so far in sing...

AI Technical Summary

Problems solved by technology

[0004] As a recognized visual problem, pose estimation has plagued researchers for many years. This is because in real life, there are often many complex scenes, and pedestrians often appear to be occluded. These scenes are also challenging scenes, but A disadvantage of using the original training set to train the network is that there is usually not enough single-person images containing occlusions to train a deep network for accurate keypoint detection / localization, good single-person pose estimation systems must Keypoint detection of individual pedestrians is robust and successful in rare and novel poses, but this problem has not been well solved

Images