CN-122023676-A - Human body three-dimensional structure modeling method based on single-station line scanning laser radar

CN122023676ACN 122023676 ACN122023676 ACN 122023676ACN-122023676-A

Abstract

The invention relates to the technical field of laser radar point cloud processing and human body posture reconstruction, in particular to a human body three-dimensional structure modeling method based on a single-station line scanning laser radar, which comprises the steps of according to the determined effective layering spacing meeting the human body characteristics in the height direction, and continuously judging the offset of the gravity center in the horizontal direction, obtaining a human body candidate region by effectively layering and clustering the human body candidate region meeting the continuity, intercepting point clouds in the height range corresponding to the human body candidate region to form a human body layering point cloud set, and determining the boundary range of the human body in the height direction. The invention does not need multi-sensor cooperation or complex learning model, and is suitable for human body three-dimensional perception and structure analysis in complex environments.

Inventors

WANG ZHUORAN
LI JIA
LAN QIUPING
LI ZIKUAN
ZHANG XINYU

Assignees

河海大学

Dates

Publication Date: 20260512
Application Date: 20260413

Claims (9)

1. A human body three-dimensional structure modeling method based on a single-station line scanning laser radar, which is characterized by comprising the following steps: S1, acquiring layered point cloud data acquired by a line scanning laser radar, respectively performing denoising processing based on adjacent point space distance statistics on each layered point cloud, and unifying a three-dimensional coordinate system by taking a laser radar emission center as an origin to obtain denoised layered point clouds with aligned coordinates; S2, extracting multi-dimensional characteristics of each layered point cloud obtained in the step S1, screening the layered point clouds based on the obtained multi-dimensional characteristics, and determining effective layering meeting human body characteristics, wherein the multi-dimensional characteristics comprise geometric characteristics, echo intensity characteristics and point cloud density characteristics; s3, continuously judging according to the determined distance between effective layering meeting the human body characteristics in the height direction and the offset of the gravity center in the horizontal direction, obtaining a human body candidate region by effective layering clustering meeting the continuity, intercepting point clouds in the height range corresponding to the human body candidate region to form a human body layering point cloud set, and determining the boundary range of the human body in the height direction; S4, aiming at each layered point cloud in the obtained human body layered point cloud set, fitting central axes of the left and right structures of the human body corresponding to the layers through iterative optimization based on point cloud symmetry characteristics, and correcting the central axis positions of abnormal layers with symmetry error exceeding a threshold value in an interpolation mode; S5, dividing the layering point cloud into left and right areas based on the central axes of the left and right structures, adaptively estimating the geometric dimension of the corresponding layering human body in the front-rear direction by combining the distribution characteristics of the left and right area point clouds, and further determining the three-dimensional center point of the corresponding layering human body; S6, carrying out continuous constraint and smooth fitting on the three-dimensional center points of the human bodies of the layers to generate a three-dimensional center axis penetrating through the height direction of the human bodies; And S7, determining the positions of key points of the human body based on the obtained three-dimensional central axis and the interlayer characteristic change of the layered point cloud, and constructing a human body skeleton model according to the human body topological relation.
2. The human body three-dimensional structure modeling method based on the single-station line scanning laser radar according to claim 1, wherein the geometric features in the S2 comprise contour spans and gravity center positions of layered point clouds in the horizontal direction; the echo intensity characteristic comprises statistical means of the echo intensities corresponding to all points of the layered point cloud respectively, and the point cloud density characteristic is the number of the point clouds in the unit space volume.
3. The human body three-dimensional structure modeling method based on the single-station line scanning laser radar according to claim 1, wherein the number of continuous effective layering corresponding to the human body candidate region in the step S3 is greater than or equal to a preset threshold.
4. The human body three-dimensional structure modeling method based on the single-station line scanning laser radar is characterized in that when the central axes of the left and right structures of the human body are determined in the step S4, the symmetrical error of the layered point cloud about the candidate central axes is calculated, the central axis positions are optimized in an iterative mode, and when the symmetrical error of a certain layered exceeds a preset error threshold, interpolation correction is conducted based on the central axis positions of adjacent layered.
5. The modeling method of the human body three-dimensional structure based on the single-station line scanning laser radar according to claim 1, wherein when the smooth fitting is performed on the human body three-dimensional center point in the step S6, the partial weighted fitting is performed by adopting a moving least square method, so that the generated human body three-dimensional center axis is kept continuous and smooth in the height direction and accords with the human body posture constraint.
6. The human body three-dimensional structure modeling method based on the single-station line scanning laser radar according to claim 1, wherein when the human body key points are located in the step S7, key point candidate layers are screened based on the interlayer difference degree of adjacent layered point clouds in the aspects of geometric form, gravity center position and echo intensity, multi-feature weighted fusion is carried out on the point clouds in the candidate layers, and the key point positions are determined.
7. The method for modeling a three-dimensional structure of a human body based on a single-station line scanning laser radar according to claim 6, wherein the inter-layer difference degree at least comprises a relative change of a profile span, a gravity center offset and an intensity change degree, and is used for quantifying change characteristics of the human body structure in a height direction.
8. The human body three-dimensional structure modeling method based on the single-station line scanning laser radar according to claim 1, wherein when the three-dimensional skeleton model is constructed in the step S7, the positions of key points are corrected based on the obtained three-dimensional central axis so that the key points are closer to the positions of real joints in a human body, and the key points are connected according to a human body topological structure to form the three-dimensional skeleton model.
9. The human body three-dimensional structure modeling method based on the single-station line scanning laser radar according to claim 1, wherein the method is used for realizing continuous modeling and posture reconstruction of a human body structure based on layered point cloud data acquired by the single-station line scanning laser radar without depending on a human body prior model, training data or multi-sensor cooperation.

Description

Human body three-dimensional structure modeling method based on single-station line scanning laser radar Technical Field The invention relates to the technical field of laser radar point cloud processing and human body posture reconstruction, in particular to a human body three-dimensional structure modeling method based on a single-station line scanning laser radar. Background With the increasing popularity of line scanning type laser radars in scenes such as industrial safety monitoring, rail transit, intelligent storage, human behavior recognition and the like, human body detection and posture estimation technology based on laser radar point cloud becomes a research hot spot. In the prior art, a camera and a deep learning model are commonly utilized to infer key points of a human body from an image, or three-dimensional laser radar point cloud is adopted to generate a human body skeleton. However, for application scenarios that are sensitive to cost, complex in environment, high in privacy requirement, and need to run in real time at the edge, the conventional technology still has obvious limitations. In an online scanning laser radar application scene, human body point cloud data are generally obtained in a layered mode along the height direction, the quantity of point clouds of each layer is limited, the spatial distribution is uneven, and a complete and continuous three-dimensional volume structure is difficult to form. Most of the existing human body detection and posture estimation methods are based on ideas such as integral point cloud clustering, single-layer contour analysis or time sequence tracking, and the like, the methods depend on relatively complete or balanced point cloud distribution conditions, and under the conditions of sparse and discrete point cloud of an online scanning laser radar, the stability of human body region identification and structural modeling is obviously limited. In the aspect of human body structure modeling, the traditional method generally estimates the positions of human body centers or key points directly based on the surface characteristics of point clouds, lacks effective constraint on space continuity between layering at different heights, is easy to generate structure jump in the height direction, and is difficult to form stable and continuous human body center structure description. In addition, the method for compensating depending on multi-frame time information has limited effect under the single-frame or real-time processing requirement, and is unfavorable for deployment in an edge end or low-computing resource environment. Therefore, in the human body sensing application facing the line scanning laser radar, a method for realizing stable positioning of a human body region and continuous modeling of a central structure by establishing a structure continuous relation in a height direction by fully utilizing the space characteristics of layered point clouds is needed, so that the reliability and the practicability of human body structure extraction are improved on the premise of not needing high-density point clouds or complex model support. Disclosure of Invention The invention aims to provide a human body three-dimensional structure modeling method based on a single-station line scanning laser radar, which aims to solve the problems in the background technology. In order to solve the technical problems, the invention provides a human body three-dimensional structure modeling method based on a single-station line scanning laser radar, which comprises the following steps: The method comprises the steps of S1, obtaining layered point cloud data collected by a line scanning type laser radar, respectively performing denoising processing based on space distance statistics of adjacent points on each layered point cloud, and unifying a three-dimensional coordinate system by taking a laser radar emission center as an origin to obtain layered point clouds with denoising and aligned coordinates, wherein the denoising processing comprises the specific processes that each point in each layered point cloud is searched for a preset number of nearest neighbors through a KD tree, the average distance between the point and the nearest neighbors is calculated, the average value and standard deviation of the average distance of all points of the layer are calculated, noise points with the average distance exceeding 2 times of standard deviation are removed, and the rule of the unified three-dimensional coordinate system is that a Z axis is in a height direction and is consistent with the layering scanning direction of the laser radar, an X axis is in a horizontal left-right direction, and a Y axis is in a front-back direction and points to one side deviating from a human body. S2, extracting multi-dimensional characteristics of each layered point cloud obtained in the step S1, screening the layered point clouds based on the obtained multi-dimensional characteristics, and de