Indoor human body detection method based on 3D (three-dimensional) point cloud image

Abstract

The invention discloses an indoor human body detection method based on a 3D (three-dimensional) point cloud image. The method includes the steps of firstly, segmentation and clustering based on the 3D point cloud image, secondly, non-human-body-region filtering and thirdly, human body region detection. According to the method, firstly, most of filtering is performed by the aid of spatial geometrical information so as to decrease a region to be processed truly to the minimum; secondly, algorithms in different subdivided fields are combined, and both a graph coloring algorithm in a graph theory and a support vector machine training method in the field of computer vision are used. By combining the algorithms, the indoor human body detection method is a universal human body detection method which is effective and low in computation complexity.

Description

technical field [0001] The invention relates to the technical field of indoor human body detection, in particular to an indoor human body detection method based on 3D point cloud images. Background technique [0002] For the technology related to the interaction between robots and people, the detection of dynamic human body is one of the core technologies. All indoor service robots, including household service robots, elderly assistance robots, shopping guide robots in shopping malls and exhibition centers, need the support of human detection function. [0003] 3D point cloud sensor, as a common sensor component for human-computer interaction and robots in recent years, can return an image composed of point clouds. Each pixel has position (x, y, z) information of the image pixel in real space. Point cloud information greatly enriches the geometric space information of images, so it is adopted by more and more applications. [0004] The present invention uses a 3D point cl...

AI Technical Summary

Problems solved by technology

[0005] Existing methods for human detection on robots or indoors usually use plane image information obtained by ordinary cameras. Most of these methods rely on background segmentation or require a huge amount of computing support.

At the same time, due to the influence of the camera due to light changes, this type of method cannot obtain a more average human detection effect in different environments.

In addition, according to the difference between the sensor configuration on the dynamic robot or the static fixed position, the detection methods are also quite different, and there is no universal and applicable rapid detection method

Images