A Human Detection Method Based on SAE Feature Visualization Learning

Abstract

The invention discloses a human body detection method based on SAE feature visualization learning, comprising: extracting image features through a CNN constructed by SAE learning according to collected depth images; visualizing the image features as a high-dimensional abstract image; The CNN constructed by the second-layer SAE learning to extract image features from the 3D abstract image; input the image features into the trained SVM classifier to obtain whether the depth image contains a human body. A method for human body detection based on SAE feature visualization learning proposed by the present invention uses depth images to extract image features and visualized features to obtain high-dimensional images. By extracting high-dimensional image features, the correct rate of human body detection is improved, and can be applied to intelligent monitoring. and human-computer interaction systems.

Description

technical field [0001] The invention relates to the technical fields of computer vision and human body detection, in particular to a human body detection method based on visual learning of SAE features. Background technique [0002] Human body detection is an important part of human motion analysis and human-computer interaction, and it is also the basic task of computer vision. It has broad application prospects in the fields of intelligent monitoring, virtual reality, human-computer interaction, and auxiliary clinical diagnosis. However, it is still a difficult problem to achieve accurate and reliable human detection in practical application scenarios because human body detection will be interfered by scene lighting changes, viewing angle changes, complex backgrounds, and pose changes. [0003] Traditional human detection methods based on color images or grayscale images, such as HoG, LBP, and Harr-like, can realize human detection in simple scenes. However, human detecti...

AI Technical Summary

Problems solved by technology

However, human detection based on color or grayscale images is greatly affected by environmental lighting, complex backgrounds, changes in human body posture, etc. When the detected object is dressed in clothes that are similar to the background color, it will also interfere with the effect of the detection algorithm.

Images