Method and device for human body detection based on embedded system

Abstract

The invention discloses a human body detection method and device based on an embedded system. The human body detection method includes: establishing a human body sample fusion channel feature set, and according to a preset algorithm for each sample in the human body sample fusion channel feature set according to The preset weight is updated iteratively at least once to obtain a human detection classifier, the classifier includes at least one sub-classifier, and the sub-classifier includes: a histogram lookup table, a feature dimension number and a threshold corresponding to the sub-classifier; obtain the image to be detected RGB color space information, determine the LUV feature information in the fusion channel feature of the image to be detected according to the pre-established RGB and LUV color space mapping relationship table; distinguish the fusion channel feature information of the image to be detected according to the classifier to determine Detect if an image contains a human body. By establishing the color space mapping relationship table of RGB and LUV, the speed of training and detection is optimized, and the efficiency of human body detection in embedded systems is improved.

Description

technical field [0001] The invention relates to the field of vision, in particular to a method and device for human body detection based on an embedded system. Background technique [0002] Human detection is widely used in monitoring systems, airport security, robotics, human-computer interaction and other fields. At present, the main human detection methods are divided into three categories: DPM (Deformable Part Model), deep network and decision forest. [0003] DPM was proposed by Felzenszwalb. As its name says, the deformable component model is a component-based detection algorithm, because it needs to design and calculate multiple local components for various parts of the human body (head, arm, etc.) during training. template, so the feature calculation is large and the detection speed is slow. [0004] Deep network is a method of using convolutional neural network to learn and train a large number of samples to achieve target detection, which can achieve end-to-end d...

AI Technical Summary

Problems solved by technology

[0003] DPM was proposed by Felzenszwalb. As its name says, the deformable component model is a component-based detection algorithm, because it needs to design and calculate multiple local components for various parts of the human body (head, arm, etc.) during training. template, so the feature calculation is large and the detection speed is slow

Images