CN-121982088-A - Polarization information guiding-based semi-global matching binocular depth estimation method

CN121982088ACN 121982088 ACN121982088 ACN 121982088ACN-121982088-A

Abstract

The invention provides a polarization information guiding-based semi-global matching binocular depth estimation method, which comprises the steps of obtaining a focus-division plane polarization optical image of a target under a binocular angle, calculating polarization information of the focus-division plane polarization optical image to obtain polarization degree information and polarization angle information, constructing a polarization matching cost function by adopting a window matching method based on the polarization angle information, constructing a gray matching cost function by adopting an SAD-Census mixed calculation method based on the focus-division plane polarization optical image, taking the polarization degree information as fusion weight, carrying out weighted fusion processing on the polarization matching cost function and the gray matching cost function to obtain a fusion cost function, carrying out semi-global matching SGM based on polarization cue guiding by adopting the fusion cost function to obtain final parallax information of the target, and carrying out three-dimensional reconstruction processing on the target by adopting final parallax information of the target to obtain target point cloud data. Based on this, a significant improvement in matching accuracy and robustness is achieved.

Inventors

LI XUAN
LI DACHUAN
SHAO XIAOPENG
LU XIAOLONG
CAI YUDONG
Pan Cunying

Assignees

西安电子科技大学

Dates

Publication Date: 20260505
Application Date: 20260126

Claims (10)

1. The semi-global matching binocular depth estimation method based on polarization information guidance is characterized by comprising the following steps of: Acquiring a focal plane polarization optical image of a target under a binocular viewing angle, and calculating polarization information of the focal plane polarization optical image to obtain polarization degree information and polarization angle information; Constructing a polarization matching cost function by adopting a window matching method based on the polarization angle information; based on the focal plane polarization optical image, adopting an SAD-Census mixed calculation method to construct a gray matching cost function; taking the polarization degree information as a fusion weight, and carrying out weighted fusion processing on the polarization matching cost function and the gray matching cost function to obtain a fusion cost function; performing semi-global matching SGM based on polarization cue guidance by adopting the fusion cost function to obtain final parallax information of a target; And carrying out three-dimensional reconstruction processing on the target through the final parallax information of the target to obtain target point cloud data.
2. The polarization information guiding-based semi-global matching binocular depth estimation method of claim 1, wherein the obtaining a focal plane polarized optical image of a target under binocular vision angle and calculating polarization information of the focal plane polarized optical image, obtaining polarization degree information and polarization angle information, comprises: acquiring a focal plane polarized optical image of a target under a binocular viewing angle; and respectively calculating the polarization degree and the polarization angle of the focal plane-separated polarized optical image according to a Stokes vector equation, and correspondingly obtaining the polarization degree information and the polarization angle information.
3. The polarization information guided semi-global matching binocular depth estimation method of claim 1, wherein the polarization matching cost function is expressed as; ; Wherein, the Representing the center point of the matching window of the split focal plane polarized optical left image, Represents the center point of the matching window of the split focal plane polarized optical right image, In the process of representing matching And The parallax between the two is not limited to the parallax, Expressed in parallax Lower part(s) The corresponding polarization matches the cost function, In the process of representing matching Is a binary code for the spreading of the characteristic window of (a), In the process of representing matching Is a binary code for the spreading of the characteristic window of (a), Representing the hamming distance.
4. A polarization information-based guided semi-global matching binocular depth estimation method according to claim 3, wherein the feature window of the center point of the split focal plane polarization optical image matching window is expressed as: ; Wherein, the Representing the center point of the matching window of the split focal plane polarized optical left image, Representation of Is used for the neighbor point of (a), Representation of And A characteristic window in between and a characteristic window in between, Representation of Is provided with a polarization angle information of (a), Representation of Is a polarization angle information of the light source.
5. The polarization information guided semi-global matching binocular depth estimation method of claim 1, wherein the fusion cost function is expressed as: ; Wherein, the Representing the center point of the matching window of the split focal plane polarized optical left image, In the process of representing matching And The parallax between the two is not limited to the parallax, Represents the center point of the matching window of the split focal plane polarized optical right image, Expressed in parallax Lower part(s) The corresponding fusion cost function is used to determine, Representation of Is used for the fusion of the texture intensity, Expressed in parallax Lower part(s) The corresponding gray level matches the cost function, Expressed in parallax Lower part(s) The corresponding polarization matches the cost function, ; Wherein, the Representing normalized polarization degree information after normalization processing is performed on the polarization degree information, Representing the pair of Values normalized by corresponding regional gradient results, the The corresponding regional gradient results are calculated based on Sobel operators, Representing a first adjusted weight of the first set of weights, Representing a second adjustment weight.
6. The polarization information guided semi-global matching binocular depth estimation method of claim 1, wherein the final disparity information of the object is represented as: ; ; Wherein, the The final parallax information representing the object is displayed, Indicating that all of the aggregate paths are present, Represent the first The path of the aggregation is a path of aggregation, Representation of The corresponding maximum parallax value is taken as, In the process of representing matching And The parallax between the two is not limited to the parallax, , Representing the center point of the matching window of the split focal plane polarized optical left image, In the process of representing matching And The parallax between the two is not limited to the parallax, Expressed in parallax Lower part(s) The corresponding aggregate cost is used to determine the aggregate cost, Expressed in parallax Lower part(s) The corresponding fusion cost function is used to determine, Representing in addition to All but the possible disparities, , Representation of In the aggregate path Lower and upper The parallax of the former point is consistent as At the cost of the aggregation at the time of the time, Representation of In the aggregate path Lower and upper The aggregation cost when the parallax of the previous point differs by 1, Representation of In the aggregate path Lower and upper The aggregate cost when the disparity of the previous point is inconsistent, Indicating under polarization guidance In the aggregate path The slow smooth penalty term of the lower part, Indicating under polarization guidance In the aggregate path The jump below smoothes the penalty term.
7. The polarization information guiding-based semi-global matching binocular depth estimation method of claim 1, wherein the performing three-dimensional reconstruction processing on the target through the final parallax information of the target to obtain target point cloud data comprises: Calculating to obtain target depth data based on the final parallax information of the target; And under world coordinates of shooting equipment, restoring the target depth data to obtain the target point cloud data.
8. The polarization information guided semi-global matching binocular depth estimation method of claim 7, wherein the calculating the target depth data based on the target's final parallax information comprises: Based on the final parallax information of the target, calculating to obtain the target depth data by adopting a binocular depth calculation method of polar line correction; the target depth data is expressed as: ; Wherein, the The depth data of the object is represented, A baseline of the photographing apparatus is indicated, Representing the lateral pixel focal length of the photographing apparatus, Representing the final parallax information of the object.
9. The semi-global matching binocular depth estimation device based on polarization information guidance is characterized by comprising an acquisition unit, a function construction unit, a fusion unit, a matching unit and a three-dimensional reconstruction unit; the acquisition unit is used for acquiring a focal plane polarization optical image of a target under a binocular vision angle, and calculating polarization information of the focal plane polarization optical image to obtain polarization degree information and polarization angle information; The function construction unit is used for constructing a polarization matching cost function by adopting a window matching method based on the polarization angle information, and constructing a gray matching cost function by adopting an SAD-Census mixed calculation method based on the focal plane-separated polarized optical image; The fusion unit is used for taking the polarization degree information as fusion weight, and carrying out weighted fusion processing on the polarization matching cost function and the gray matching cost function to obtain a fusion cost function; The matching unit is used for performing semi-global matching SGM based on polarization cue guidance by adopting the fusion cost function to obtain final parallax information of a target; and the three-dimensional reconstruction unit is used for carrying out three-dimensional reconstruction processing on the target through the final parallax information of the target to obtain target point cloud data.
10. A polarization information based guided semi-global matching binocular depth estimation device comprising a processor, a storage medium storing machine readable instructions executable by the processor, the processor in communication with the storage medium via the bus when the polarization information based guided semi-global matching binocular depth estimation device is in operation, the processor executing the machine readable instructions to perform the steps of the polarization information based guided semi-global matching binocular depth estimation method of any one of claims 1-8.

Description

Polarization information guiding-based semi-global matching binocular depth estimation method Technical Field The invention relates to the technical field of machine vision, in particular to a semi-global matching binocular depth estimation method based on polarization information guiding. Background With the rapid development of computer vision and machine vision technologies, three-dimensional perception technologies based on visual information are widely applied to numerous fields such as robot navigation, automatic driving, industrial detection and augmented reality. The key to realizing three-dimensional scene understanding and space measurement is accurate acquisition of depth information. In many depth acquisition technologies, the binocular stereoscopic vision method becomes an important choice in an actual engineering system due to the advantages of low hardware cost, no need of an active light source, strong adaptability and the like. According to the method, parallax is calculated by carrying out pixel matching on images acquired from different visual angles, and then the three-dimensional structure of a scene is restored. However, this process has extremely high requirements on matching accuracy and stability, and how to achieve robust and accurate matching is a core technical bottleneck currently faced, especially when faced with practical application environments such as complex illumination, weak textures, or the existence of reflections. In order to solve the matching problem, a plurality of stereo matching methods are proposed in the prior art. The binocular stereoscopic vision method based on semi-global matching is a mature and widely adopted technical scheme in engineering practice. The method generally builds pixel-level matching cost according to gray level or color information of an image after finishing image epipolar correction, and performs cost aggregation along a plurality of preset directions, and simultaneously introduces smoothness constraint to optimize continuity and consistency of a parallax map. The method has better balance between calculation efficiency and result stability, so that the method has higher practicability in the fields of vehicle-mounted vision, three-dimensional measurement and the like. Nevertheless, existing stereo matching techniques still suffer from significant drawbacks. First, the conventional method represented by semi-global matching is highly dependent on the intensity information (e.g., gray scale, color) of the image. When weak textures and uniform color areas exist in a scene or are disturbed by intense illumination changes and high light reflection, the intensity information lacks sufficient discrimination and stability, and mismatching or parallax holes are easily caused. Secondly, the method generally adopts isotropic smooth constraint, is difficult to adaptively adjust according to the actual geometric structure of the scene, and easily causes loss of depth details. In addition, the stereo matching method based on deep learning, which is emerging in recent years, is high in accuracy on certain data sets, but depends on large-scale labeling data for training, has insufficient generalization capability, and is difficult to meet the severe requirements of practical engineering application, especially embedded or real-time systems in terms of reliability, interpretability and computing resource consumption in complex scenes. Therefore, the precision and the robustness of the current technology are still to be further improved when the current technology is used for dealing with challenging scenes such as weak textures, complex reflections and the like. Disclosure of Invention In order to solve the problems in the prior art, the invention provides a semi-global matching binocular depth estimation method based on polarization information guiding. The technical problems to be solved by the invention are realized by the following technical scheme: in a first aspect, the present invention provides a polarization information guiding-based semi-global matching binocular depth estimation method, including: Acquiring a focal plane polarization optical image of a target under a binocular viewing angle, and calculating polarization information of the focal plane polarization optical image to obtain polarization degree information and polarization angle information; constructing a polarization matching cost function by adopting a window matching method based on the polarization angle information; based on the polarized optical image of the focal plane, adopting an SAD-Census mixed calculation method to construct a gray matching cost function; taking the polarization degree information as fusion weight, and carrying out weighted fusion processing on the polarization matching cost function and the gray matching cost function to obtain a fusion cost function; performing semi-global matching SGM based on polarization cue guidance by adopting a fusion cost fun