Shielded object reasoning type search method based on grid discrete representation containability analysis method

Abstract

The invention belongs to the technical field of robot target search, discloses an occluded object reasoning type search method based on a grid discrete representation containability analysis method, and aims to solve the defects of poor generalization, high calculation complexity, no consideration of object placement constraint and the like when an occluded object reasoning problem is processed in the prior art. According to the method, geometric information of a target object and a container is discretized into grid representation, and accommodation analysis is carried out in combination with support surface constraint and height constraint of object placement, so that a potential position where the shielded target object may exist is quickly and accurately inferred. The method has the advantages of being high in universality, high in calculation efficiency, high in accuracy, free of a large amount of labeled data, high in interpretability and the like, and is suitable for target object search tasks in various sheltered scenes.

Description

Technical Field

[0001] The present invention belongs to the technical field of robot target search, and particularly relates to an inference-based search method for occluded objects based on a grid discrete representation and admissibility analysis method. Background Art

[0002] In the robot target search task, the target object may be occluded by other objects, increasing the difficulty of search. The existing target search methods are mainly divided into two categories: learning-based occupancy probability distribution methods and three-dimensional information completion-based methods. However, these methods have the following problems:

[0003] Poor generalization: Learning-based methods need to construct a large number of labeled datasets for specific scenarios. When the scenario changes (such as object types, container types, lighting conditions, etc.), the model needs to be retrained, and the generalization ability is poor.

[0004] High computational complexity: Three-dimensional information completion-based methods need to perform complex three-dimensional reconstruction and optimization, with a large amount of calculation, and it is difficult to meet the real-time requirement.

[0005] Do not consider object placement constraints: Existing methods rarely consider the physical constraints of object placement in the real world (such as support surfaces, height limits, etc.), resulting in potential positions deduced may not conform to the actual situation. Summary of the Invention

[0006] The purpose of the present invention is to provide an inference-based search method for occluded objects based on a grid discrete representation and admissibility analysis method to solve the above technical problems.

[0007] To solve the above technical problems, the specific technical solution of an inference-based search method for occluded objects based on a grid discrete representation and admissibility analysis method of the present invention is as follows:

[0008] An inference-based search method for occluded objects based on a grid discrete representation and admissibility analysis method, comprising the following steps:

[0009] Step 1: Image acquisition and preprocessing: The robot uses a camera to acquire the RGB image inside the container that may accommodate the object and performs preprocessing;

[0010] Step 2: Visible object detection and segmentation: Use an image segmentation algorithm to segment the visible objects in the image and identify their categories;

[0011] Step 3: Grid Discrete Representation: When an object is placed, the contact surface between the object and other objects is fixed, i.e., the support surface; when an object is placed in a container, the surface that bears the placed object is usually fixed, i.e., the bearing surface; project the bearing surface of the openable container and the support surface of the target object onto a two-dimensional plane and discretely represent them using grids of a set size, where each grid represents a small area;

[0012] Step 4: Calculation of the Occupied Area of Visible Objects: According to the segmentation result, calculate the occupied area of each visible object in the grid, mark the grids occupied by the bearing surface of the container as occupied states, and the unoccupied grids as candidate positions; convert the size of the support surface of the target object into the size in terms of the number of grids;

[0013] Step 5: Accommodation Analysis: For each unoccupied grid, conduct accommodation analysis; Step 6: Sorting of Potential Positions: Sort all unoccupied grids according to the results of the accommodation analysis;

[0014] Step 7: Inference Search: Check each potential position in turn according to the sorting result;

[0015] Step 8: Target Object Localization: When the target object is found at a certain potential position, the search ends. Further, Step 2 includes extracting the three-dimensional point cloud information of visible objects by combining depth images. After obtaining the point cloud of visible objects in the scene, the approximate structure, position, and pose information of the visible objects can be obtained by fitting with bounding boxes.

[0016] Further, the accommodation analysis in Step 5 screens candidate accommodation objects in two stages. First, in the stage of selecting an openable container that may store the target object; second, after opening the openable container, judge in which space blocked by visible objects the target object may be located; use an OBB to represent the target object to be searched, and the accommodation analysis problem is simplified to a problem of whether a cuboid of a given size can be placed in a polyhedron. The container in the container selection stage of the openable container is represented by a cuboid; the container in the visible object selection stage is represented by a frustum. When an object is placed, the contact surface between the object and other objects is fixed, i.e., the support surface, and when an object is placed in a container, the surface that bears the placed object is usually fixed, i.e., the bearing surface. Therefore, the accommodation problem of the target object in a specific container is transformed into the accommodation problem of the relative height of the object with respect to the bearing surface and the accommodation problem of the support surface of the target object in the bottom surface of the container.

[0017] Further, Step 5 includes the following steps:

[0018] Step 5.1: Accommodation Analysis Based on the Support Surface:

[0019] Step 5.2: Accommodation Analysis Based on Height Limitations;

[0020] Step 5.3: Analysis of the accommodatability combining the support surface and height constraints:

[0021] Furthermore, for the openable container in Step 5.1, based on the gridded discrete representation, by detecting the number of consecutive unoccupied grids in the X-axis and Y-axis directions in the container bearing surface area, it is determined whether there is an area where the number of consecutive unoccupied grids is greater than or equal to the grid number size of the target object. The bearing surface of the openable container is described by a two-dimensional rectangle. The conditions for the target object to be accommodated in the openable container are: the height of the target object is lower than the height of the internal space of the openable container, and the support surface of the target object bounding box can be accommodated within the rectangle of the openable container bearing surface; if there is such an area, it is determined that the support surface of the target object can be accommodated within the container bearing surface; if it cannot be accommodated in the current direction, the container grid surface is rotated by a certain angle, and the above detection process is repeated until the cumulative rotation angle reaches the preset threshold; if there is still no accommodatable solution, it is determined that the target object cannot be accommodated in the container.

[0022] Furthermore, in Step 5.1, the bottom geometries of the target object and the container are discretized into grids using the gridding method, and the accommodatability between the two is analyzed based on the grid representation. The implementation steps are as follows: Step 5.1.1: Use the container bottom information to construct a two-dimensional plane, set the grid size, and convert the plane into a grid surface;

[0023] Step 5.1.2: Traverse all grids, set the grid value occupied by the container bearing surface to 1, and otherwise set it to 0 to determine the grid area available for placing the target item;

[0024] Step 5.1.3: Convert the support surface size of the target object into the grid number size using the grid size; detect the number of consecutive grids in the x-axis and y-axis directions in the container bearing surface area. If the number of consecutive grids at a certain position is greater than the grid number size of the target item, the support surface of the target object can be accommodated within the container bottom;

[0025] Step 5.1.4: If the support surface of the target item cannot be accommodated within the container bottom in this analysis, rotate the container grid surface by a certain angle, and repeat Step 5.1.2 and Step 5.1.3. When the cumulative rotation angle reaches 90° and there is still no accommodatable solution, the target object cannot be accommodated in the container.

[0026] Furthermore, for the occlusion space formed by visible objects in Step 5.2, based on the perspective projection principle, calculate the occlusion space formed by the visible object on the support surface; determine whether the height of the target object is lower than the height of the occlusion space at a certain position when the target object is located in the occlusion space; for the area that meets the height limit, perform the accommodatability analysis based on the support surface described in Step 5.1 to obtain the possible positions of the target object in the occlusion space.

[0027] Further, step 5.2 includes the following steps:

[0028] Assume that the position of the viewpoint in the scene is l, and the equation of the projection plane is defined as The coordinate of a point on the visible object is p, and the relationship between point p and its projection p' on the projection plane can be expressed as: p' = M·p

[0029] where the matrix M is: After obtaining the projection of the visible object on the projection plane, the corresponding occlusion space is obtained;

[0030] When it is determined that the target object is located somewhere in the occlusion space, whether its height is lower than the height of the occlusion space at that position: First, generate the shadow formed by the visible object on the support surface, and then, let the shadow area S shadow The height of the inner points relative to the support surface is the height of the target object, and a new point cloud S shadow is obtained. If the projection of a point in S' shadow at this viewing angle is still within S shadow inside, then the corresponding point of this point in S shadow may accommodate the target item; perform an accommodation analysis within the plane of the area that meets the height limit, and the possible positions of the target item in the scene can be obtained.

[0031] Further, step 5.3 comprehensively uses the accommodation analysis methods described in step 5.1 and step 5.2, decomposes the accommodation problem of the target object in a specific container into the accommodation problem of the object's height relative to the bearing surface and the accommodation problem of the support surface of the target object in the bottom surface of the container, and determines the possibility and potential position of the target object through the combination of the two.

[0032] Further, the basis for the sorting in step 6 is the stability of the placement of the target object on the grid and the positional relationship between the grid and other known objects.

[0033] A method for reasoning and searching occluded objects based on grid discrete representation and accommodation analysis of the present invention has the following advantages:

[0034] 1. Strong generality and wide application range: This method does not depend on the data set of a specific scene, does not require model training for different scenes, and is applicable to the internal environments of various different openable containers, with strong generality and adaptability.

[0035] 2. High computational efficiency and good real-time performance: This method mainly performs grid-based discrete representation and simple geometric calculations on a two-dimensional plane, avoiding complex three-dimensional reconstruction and optimization processes, greatly reducing the computational amount, and being able to meet the requirements of real-time decision-making of robots.

[0036] 3. High accuracy and strong robustness: This method fully considers the physical constraints of object placement in the real world, namely the support surface and height limitations, making the potential positions of the target objects deduced more in line with the actual situation and avoiding unreasonable prediction results.

[0037] 4. No need for a large amount of labeled data: This method does not require any training data, avoiding the cost and time of labeling data and enabling rapid deployment and application.

[0038] 5. Strong interpretability: This method is based on clear geometric rules and physical constraints, and the reasoning process is clear and transparent, making it easy to understand and debug. Description of the Drawings

[0039] Figure 1 : Schematic diagrams of two accommodation detection scenarios.

[0040] Figure 2 : Schematic diagram of accommodation detection.

[0041] Figure 3 : Schematic diagram of accommodation detection based on grid discretization.

[0042] Figure 4 : Schematic diagram of the formation of the occlusion space.

[0043] Figure 5 : Schematic diagram of extracting the accommodation area based on height limitations. Detailed Implementation Manner

[0044] To better understand the purpose, structure, and function of the present invention, the following further describes in detail a method for reasoning and searching for occluded objects in an accommodation analysis method based on grid discrete representation of the present invention with reference to the accompanying drawings.

[0045] The present invention proposes a method for reasoning and searching for occluded objects in an accommodation analysis method based on grid discrete representation, specifically including the following steps:

[0046] Step 1: Image acquisition and preprocessing: The robot uses a camera to acquire the RGB image inside the container that may accommodate objects and performs preprocessing, such as image denoising, color correction, etc.

[0047] Step 2: Visible object detection and segmentation: Use an image segmentation algorithm (e.g., Mask R-CNN) to segment the visible objects (including partially visible objects) in the image and identify their categories. Combine the depth image to extract the three-dimensional point cloud information of the visible objects. After obtaining the point cloud of the visible objects in the scene, use a bounding box fitting to obtain the approximate structure, position, and pose information of the visible objects.

[0048] Step 3: Grid discrete representation: AsFigure 1 As shown. When placing an object in real life, the contact surface between the object and other objects is fixed, that is, the support surface. When placing an object in a container, the surface that bears the placed object is usually fixed, that is, the bearing surface. Project the bearing surface of the openable container and the support surface of the target object onto a two-dimensional plane and represent them discretely using a grid with a set size. Each grid represents a small area.

[0049] Step 4: Calculation of the occupied area of the visible object: According to the segmentation result, calculate the occupied area of each visible object in the grid. Mark the grids occupied by the bearing surface of the container as occupied, and the unoccupied grids as candidate positions; convert the size of the support surface of the target object into the size in terms of the number of grids.

[0050] Step 5: Feasibility analysis: For each unoccupied grid, conduct a feasibility analysis. The feasibility analysis can screen candidate objects to be accommodated in two stages. First, the stage of selecting an openable container that may store the target object. Second, after opening the openable container, determine in which space blocked by visible objects the target object may be located. Use an OBB to represent the target object to be searched, and the feasibility analysis problem is simplified to the problem of whether a cuboid with a given size can be placed into a polyhedron. The container in the stage of selecting the openable container is represented by a cuboid; the container in the stage of selecting visible objects is represented by a frustum. The feasibility problem of the target object in a specific container is transformed into the feasibility problem of the height of the object relative to the bearing surface and the feasibility problem of the support surface of the target object in the bottom surface of the container.

[0051] Step 5.1: Feasibility analysis based on the support surface: For the openable container, based on the above-mentioned grid-based discrete representation, by detecting the number of continuously unoccupied grids in the X-axis and Y-axis directions in the area of the bearing surface of the container, determine whether there is an area where the number of continuously unoccupied grids is greater than or equal to the size of the target object in terms of the number of grids. Use a two-dimensional rectangle to describe the bearing surface of the openable container. The conditions for the target object to be accommodated in the openable container are: the height of the target object is lower than the height of the internal space of the openable container, and the support surface of the bounding box of the target object can be accommodated within the rectangle of the bearing surface of the openable container. If there is such an area, it is determined that the support surface of the target object can be accommodated in the bearing surface of the container; if it is not feasible in the current direction, rotate the grid surface of the container by a certain angle and repeat the above detection process until the cumulative rotation angle reaches a preset threshold; if there is still no feasible solution, it is determined that the target object cannot be accommodated in the container.

[0052] Use the grid method to discretize the bottom geometries of the target object and the container into grids, and analyze the feasibility between the two based on the grid representation. The implementation steps are as follows:

[0053] Step 5.1.1: Construct a two-dimensional plane using the bottom surface information of the container, set the grid size, and convert the plane into a grid surface;

[0054] Step 5.1.2: Traverse all grids, set the grid values occupied by the container bearing surface to 1, and set the others to 0 to determine the grid area available for placing the target item;

[0055] Step 5.1.3: Convert the support surface size of the target object into the grid number size using the grid size; Detect the continuous grid numbers in the x-axis and y-axis directions of the container bearing surface area. If the continuous grid number at a certain position is greater than the grid number size of the target item, the support surface of the target object can be accommodated on the bottom surface of the container, as Figure 2 shown;

[0056] Step 5.1.4: If the support surface of the target item cannot be accommodated on the bottom surface of the container in this analysis, rotate the container grid surface by a certain angle, and repeat Step 5.1.2 and Step 5.1.3. When the cumulative rotation angle reaches 90° and there is still no accommodatable solution, the target object cannot be accommodated in this container. An example of the accommodation detection diagram is as Figure 3 shown.

[0057] Step 5.2: Accommodation analysis based on height limit;

[0058] For the occlusion space formed by visible objects, based on the principle of perspective projection, calculate the occlusion space formed by the visible objects on the support surface; Determine whether the height of the target object is lower than the height of the occlusion space at a certain position when the target object is located in the occlusion space; For the area that meets the height limit, perform the accommodation analysis based on the support surface described in Step 5.1 to obtain the possible positions of the target object in the occlusion space. The occlusion space of visible objects is similar to the shadow volume of the object under a point light source. Assume that the position of the viewpoint in the scene is l, and the equation of the projected plane is defined as The coordinate of a certain point on the visible object is p, and the relationship between point p and its projection p' on the projected plane can be expressed as: p' = M·p

[0059] where the matrix M is:

[0060] After obtaining the projection of the visible object on the projected plane, the corresponding occlusion space can be obtained. As Figure 4 shown, S is the camera position, ΔABC is the visible object in the camera, ΔA'B'C' is the area of the plane Γ occluded by the object, and the polyhedron ABCC'B'A' is the occlusion space formed by ΔABC.

[0061] Different from an openable container, the height of the occlusion space of a visible object varies relative to the support surface. When it is necessary to determine the height of a target object at a certain position in the occlusion space, it is necessary to determine whether its height is lower than the height of the occlusion space at that position. As Figure 5 shown, first generate the shadow formed by the visible object on the support surface, and then let the height of the points inside the shadow area S shadow relative to the support surface be the height of the target object, obtaining a new point cloud S shadow . If the projection of a point in S' shadow at this viewing angle is still within S shadow , then the corresponding point of this point in S shadow may accommodate the target item. The possible positions of a target item with a height of 0.05 m in the scene shown in Figure 5 (a) are shown in Figure 5 (b).

[0062] Perform an accommodation analysis on the plane of the area that meets the height constraint, and the possible positions where the target item may exist in the scene can be obtained.

[0063] Step 5.3: Accommodation analysis combining the support surface and height constraints: Comprehensively apply the accommodation analysis methods described in Step 5.1 and Step 5.2, decompose the accommodation problem of the target object in a specific container into the accommodation problem of the object's height relative to the bearing surface and the accommodation problem of the target object's support surface in the bottom surface of the container. By combining the two, determine the possibility and potential positions of the target object.

[0064] Step 6: Sorting of potential positions: According to the results of the accommodation analysis, sort all unoccupied grids. The basis for sorting can be the stability of placing the target object on this grid, the positional relationship between this grid and other known objects, etc.

[0065] Step 7: Inference-based search: Check each potential position in turn according to the sorting result. The checking method can be to directly observe this position or use a robotic arm to detect this position.

[0066] Step 8: Target object positioning: When the target object is found at a certain potential position, the search ends.

[0067] This algorithm not only relies on visual information but also combines the spatial attributes of the item and potential position reasoning, greatly improving the recognition rate of occluded items and optimizing the search path.

[0068] Example:

[0069] Scene description: There are several items placed in a drawer, including a target object (e.g., a screwdriver), and some other occluding objects (e.g., pliers, wrenches, measuring tapes, etc.). The robot needs to find the screwdriver through visual observation and reasoning without knowing its specific location.

[0070] Step 1: Image acquisition and preprocessing: The robot uses a camera to obtain the RGB image inside the drawer and performs preprocessing, such as image denoising, color correction, etc.

[0071] Step 2: Visible object detection and segmentation: Use an image segmentation algorithm (e.g., Mask R-CNN) to segment the visible objects (including partially visible objects) in the image and identify their categories (e.g., pliers, wrenches, measuring tapes, etc.).

[0072] Step 3: Grid discrete representation: Discretize the bottom plane of the drawer into a two-dimensional grid. Each grid represents a small area.

[0073] Step 4: Occupied area calculation of visible objects: According to the segmentation result, calculate the occupied area of each visible object in the grid. Mark the grids occupied by the object as "occupied".

[0074] Step 5: Feasibility analysis: For each unoccupied grid, conduct a feasibility analysis:

[0075] Step 5.1: Support surface judgment: Judge whether there is a support surface below the grid. For example, if the bottom of the grid is the bottom of the drawer or the top of other objects, it is considered that there is a support surface.

[0076] Step 5.2: Height limit: Judge whether there is enough space above the grid to accommodate the target object. According to the height of the target object and the height of the current grid, judge whether the target object can be placed on this grid.

[0077] Step 6: Sorting of potential positions: According to the results of the feasibility analysis, sort all unoccupied grids. The sorting basis can be the stability of placing the target object on this grid, the positional relationship between this grid and other known objects, etc.

[0078] Step 7: Inference-based search: Check each potential position in turn according to the sorting result. The checking method can be to directly observe this position or use the robotic arm to detect this position.

[0079] Step 8: Target object localization: When the target object is found at a certain potential position, the search ends.

[0080] It will be understood that the present invention is described by way of some embodiments, and those skilled in the art will appreciate that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the present invention. Additionally, under the teachings of the present invention, these features and embodiments can be modified to adapt to specific circumstances and materials without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application belong to the scope protected by the present invention.

Claims

1. A method for searching for occluded objects based on the grid discrete representation accommodation analysis method, characterized in that: The following steps are involved: Step 1: Image acquisition and preprocessing: The robot uses a camera to acquire an RGB image of the interior of a container that may contain an object and performs preprocessing. Step 2: Visual object detection and segmentation: Use image segmentation algorithms to segment visible objects in the image and identify their categories; Step 3: Grid discrete representation: When an object is placed, the contact surface between the object and other objects is fixed, that is, the supporting surface; when an object is placed in a container, the surface that supports the placed object is usually fixed, that is, the bearing surface; the bearing surface of the openable container and the supporting surface of the target object are projected onto a two-dimensional plane and discretized using a grid of a set size, where each grid represents a small area; Step 4: Calculation of the occupied area of ​​the visible object: Based on the segmentation results, calculate the occupied area of ​​each visible object in the grid, mark the grid occupied by the container bearing surface as occupied, and mark the unoccupied grid as a candidate position; Convert the support surface size of the target object into the size of the number of grids; Step 5: Accommodability analysis: For each unoccupied grid, perform an accommodation analysis; Step 6: Potential location sorting: sort all unoccupied grids according to the results of the accommodation analysis; Step 7: Inferential search: Check each potential location in turn according to the sorted results; Step 8: Target object location: The search ends when the target object is found at a potential location.

2. The inference-based search method for occluded objects based on the grid discrete representation accommodation analysis method according to claim 1, characterized in that: The step 2 includes extracting three-dimensional point cloud information of visible objects in combination with the depth image. After obtaining the point cloud of the visible objects in the scene, the approximate structure, position and posture information of the visible objects can be obtained by using bounding box fitting.

3. The inference-based search method for occluded objects based on the grid discrete representation accommodation analysis method according to claim 1, characterized in that: The containment analysis of step 5 screens candidate containment objects in two stages: first, selecting an openable container that may store the target object; second, after opening the openable container, determining in which visible objects the target object may be located in the space blocked; using OBB to represent the target object to be searched, the containment analysis problem is simplified to the question of whether a cuboid of a given size can fit into a polyhedron, and the container in the openable container selection stage is represented by a cuboid; The container in the visual object selection stage is represented by a frustum. When an object is placed, the contact surface between the object and other objects is fixed, that is, the supporting surface. When an object is placed in a container, the surface that supports the placed object is usually fixed, that is, the bearing surface. Therefore, the problem of the containment of the target object in a specific container is transformed into the problem of the containment of the object relative to the bearing surface height, and the problem of the containment of the target object support surface in the bottom surface of the container.

4. The inference-based search method for occluded objects based on the grid discrete representation accommodation analysis method according to claim 3, characterized in that: The step 5 comprises the following steps: Step 5.1: Capacity analysis based on the supporting surface: Step 5.2: Accommodability analysis based on height restrictions; Step 5.3: Accommodability analysis combining support surface and height constraints.

5. The inference-based search method for occluded objects based on the grid discrete representation accommodation analysis method according to claim 4, characterized in that: In step 5.1, for an openable container, based on the grid discrete representation, by detecting the number of continuous unoccupied grids in the container bearing surface area in the X-axis and Y-axis directions, it is determined whether there is an area with a continuous number of unoccupied grids greater than or equal to the number of grids of the target object, and the bearing surface of the openable container is described using a two-dimensional rectangle. The condition that the target object can be accommodated in the openable container is that: the height of the target object is lower than the height of the internal space of the openable container, and the support surface of the target object bounding box can be accommodated in the rectangle of the bearing surface of the openable container; if so, it is determined that the support surface of the target object can be accommodated in the bearing surface of the container; If the container cannot be accommodated in the current direction, the container grid surface is rotated by a certain angle and the above detection process is repeated until the cumulative rotation angle reaches the preset threshold; If there is still no solution that can accommodate the object, it is determined that the target object cannot be accommodated in the container.

6. The inference-based search method for occluded objects based on the grid discrete representation accommodation analysis method according to claim 5, characterized in that: The step 5.1 uses a meshing method to discretize the target object and the bottom surface geometry of the container into meshes, and analyzes the containment between the two based on the mesh representation. The implementation steps are as follows: Step 5.1.1: Use the container bottom surface information to construct a two-dimensional plane, set the grid size, and convert the plane into a grid surface; Step 5.1.2: Traverse all grids, set the grid value occupied by the container bearing surface to 1, otherwise set it to 0, and determine the grid area that can be used to place the target object; Step 5.1.3: Use the grid size to convert the target object's support surface size into a grid number size; detect the number of continuous grids in the x-axis and y-axis directions of the container's support surface area. If the number of continuous grids at a certain position is greater than the grid number size of the target object, the target object's support surface can be accommodated on the bottom surface of the container; Step 5.1.4: If the support surface of the target object cannot be accommodated in the bottom surface of the container in this analysis, rotate the container grid surface by a certain angle, and repeat steps 5.1.2 and 5.1.

3. When the cumulative rotation angle reaches 90° and there is still no solution that can accommodate the target object, the target object cannot be accommodated in the container.

7. The inference-based search method for occluded objects based on the grid discrete representation accommodation analysis method according to claim 6, characterized in that: The step 5.2 calculates the occlusion space formed by the visible object on the supporting surface based on the perspective projection principle, with respect to the occlusion space formed by the visible object; determines whether the height of the target object is lower than the height of the occlusion space at that position when the target object is located at a certain position in the occlusion space; and performs the support surface-based accommodation analysis described in step 5.1 on the area that meets the height restriction to obtain the possible position of the target object in the occlusion space.

8. The inference-based search method for occluded objects based on the grid discrete representation accommodation analysis method according to claim 7, characterized in that: The step 5.2 comprises the following steps: Assuming the viewpoint is located at l in the scene, the equation of the projected plane is defined as The coordinates of a point on the visible object are p. The relationship between point p and its projection p' on the projected plane can be expressed as: p' = M·p The matrix M is: After obtaining the projection of the visible object on the projected plane, the corresponding occlusion space is obtained; When the target object is located somewhere in the occlusion space, whether its height is lower than the occlusion space height at that location: first generate the shadow of the visible object on the support surface, and then let the shadow area S shadow The height of the inner point relative to the support surface is the height of the target object, and the new point cloud S is obtained. shadow , if S' shadow The projection of a point in S at this viewing angle is still shadow If the point is within S shadow The corresponding points may accommodate the target object; by trying the accommodation analysis in the plane for the area that meets the height restriction, the possible location of the target object in the scene can be obtained.

9. The inference-based search method for occluded objects based on the grid discrete representation accommodation analysis method according to claim 6, characterized in that: The step 5.3 comprehensively applies the containment analysis methods described in steps 5.1 and 5.2 to decompose the containment problem of the target object in a specific container into the containment problem of the height of the object relative to the supporting surface and the containment problem of the target object's supporting surface in the bottom of the container. By combining the two, the possibility and potential location of the target object's existence are determined.

10. The inference-based search method for occluded objects based on the grid discrete representation accommodation analysis method according to claim 1, characterized in that: The basis for the sorting in step 6 is the stability of the target object placed on the grid and the positional relationship between the grid and other known objects.

Images