Human body detection method and device based on depth image

Abstract

The invention provides a human body detection method and device based on a depth image. The method comprises the steps: calculating a left-right disparity map, a Y-direction projection map and an X-direction projection map of a depth image; respectively converting the three images into a first gray level image, a second gray level image and a third gray level image, respectively using the three images as an R component image, a G component image and a B component image of the depth image to synthesize a three-channel color coding image, and inputting the three-channel color coding image into aconvolutional neural network detection model for detection. By adopting the method provided by the invention, three-channel color coding can be carried out on the depth image by utilizing the spatialposition information, and a depth convolution network for processing a traditional optical image can be directly used for a three-channel coded image without being modified without assistance of a visible light image.

Description

technical field [0001] The invention relates to the technical fields of computer vision and image processing, in particular to a method and device for detecting a human body based on a depth image. Background technique [0002] Human detection is a traditional and extremely important topic in the field of computer vision and image processing. From intelligent traffic control, security and anti-terrorism, to automatic driving, the application of human detection is inseparable. With the popularity of depth cameras, human body detection based on depth images has become a new hotspot—because depth images can overcome many shortcomings of traditional visible light images, are less affected by light, and can protect personal information. However, the most important model in computer vision, the convolutional neural network model, does not work well for depth images, because convolutional neural networks usually require input to be three-channel, while depth images have only one c...

AI Technical Summary

Problems solved by technology

[0004] In view of this, the present invention provides a human body detection method and device based on a depth image, which can use spatial position information to encode a depth image source with only one channel into a three-channel RGB image, thereby solving the most widely used problem in computer vision. The convolutional neural network directly processes the depth image, a single-channel image, which is not effective, and has high efficiency and human detection accuracy.

Images