Method, system, device and equipment for identifying non-line-of-sight object

Abstract

The invention discloses a method, a system, a device and equipment for identifying a non-line-of-sight object, and the method comprises the steps: setting sensing equipment, and enabling the set sensing equipment to directly observe the non-line-of-sight object and an area where the non-line-of-sight object is located; loading a global point cloud map of the position of the non-line-of-sight object, and calibrating the sensing equipment by adopting an automatic calibration algorithm to obtain a global pose of the sensing equipment under the global point cloud map of the position of the non-line-of-sight object; recognizing a non-line-of-sight object entering the global point cloud point diagram by adopting an automatic recognition algorithm, and calculating a local pose of the non-line-of-sight object relative to the sensing equipment; the local pose of the non-line-of-sight object is converted into a global pose through a coordinate conversion algorithm in combination with the global pose calibrated by the sensing device; and taking the turned global pose as a target point location, and carrying out an identification operation task. In practical application, the method can solve the problem of accurate positioning of the forklift AGV in a non-line-of-sight scene, and realizes automatic and high-precision execution of an unloading task.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of automated logistics, and in particular to a method, system, device and equipment for identifying non-line-of-sight objects. BACKGROUND

[0002] In the scenario of forklift AGV unloading from the truck box, since the forklift and the truck box are in different non-line-of-sight scenes, the forklift cannot directly identify the specific position of the specific shelf (or pallet) unloading point through its own perception system. To achieve this function, it is necessary to use an identification system for identification, and the identification system needs to have two capabilities, one is to identify the pallet and calculate the relative coordinate information of the shelf (or pallet) in the local visual image, and the other is to convert the local coordinates of the shelf (or pallet) to the global coordinate system unified with the AGV scheduling system as the unloading target point.

[0003] In the prior art, although local identification can be achieved through a visual model, the positioning stability and precision are insufficient due to factors such as non-line-of-sight occlusion and light changes, and there is a lack of mature global coordinate conversion scheme, which makes it difficult to efficiently complete the non-line-of-sight unloading task. Moreover, the existing identification method needs to be too complex, and needs to be set on the forklift with a model and a visual detection device, which often causes recognition errors due to light and other reasons during use, resulting in low recognition accuracy and seriously affecting the use efficiency. SUMMARY

[0004] To solve the above-mentioned technical problems in the prior art, the present application provides a method, system, device and equipment for identifying non-line-of-sight objects, which can solve the problem of accurate positioning of forklift AGV in a non-line-of-sight scene and realize automatic and high-precision execution of the unloading task.

[0005] The technical solution of the present application is as follows:

[0006] In a first aspect, a method for identifying non-line-of-sight objects is provided, comprising:

[0007] Setting a perception device, so that the set perception device can directly observe the non-line-of-sight object and the area where it is located;

[0008] Loading a global point cloud map of the location of the non-line-of-sight object, using an automatic calibration algorithm to calibrate the perception device, and obtaining the global pose of the perception device in the global point cloud map of the location of the non-line-of-sight object;

[0009] Using an automatic identification algorithm to identify the non-line-of-sight object entering the global point cloud map, and calculating the local pose of the non-line-of-sight object relative to the perception device;

[0010] The local pose of the non-line-of-sight object is converted into a global pose through a coordinate conversion algorithm combined with the global pose calibrated by the perception device.

[0011] The converted global pose is taken as a target point for performing the identification task.

[0012] In some optional implementations, the perception device includes a perception sensor.

[0013] In some optional implementations, the perception device is calibrated by using an automatic calibration algorithm to obtain a global pose of the perception device in a global point cloud map at a location of the non-line-of-sight object, including:

[0014] An initial pose parameter of the perception device is input;

[0015] The point cloud data collected by the perception device in real time is matched with the global point cloud map through the automatic calibration algorithm;

[0016] The initial pose parameter is iteratively optimized until a matching error is less than a preset threshold.

[0017] In some optional implementations, the preset threshold includes 5 cm.

[0018] In some optional implementations, the automatic calibration algorithm includes a point cloud global positioning matching method.

[0019] In some optional implementations, the automatic identification algorithm includes a deep learning model, training and input data of the deep learning model include images and point cloud data of the non-line-of-sight object under multiple angles and multiple light conditions, and an output of the deep learning model is a local pose of the non-line-of-sight object.

[0020] In some optional implementations, the local pose of the non-line-of-sight object is converted into a global pose through a coordinate conversion algorithm combined with the global pose of the perception device by inverse rotation matrix operation calculation, wherein the inverse rotation matrix operation includes:

[0021]

[0022] wherein, represents a local pose of the non-line-of-sight object relative to the perception device, represents a global pose of the perception device, represents a local pose of the non-line-of-sight object.

[0023] In a second aspect, a system for identifying a non-line-of-sight object is provided, including:

[0024] An observation module, configured to enable a perception device to directly observe the non-line-of-sight object and a region where the non-line-of-sight object is located.

[0025] A global pose calibration module is configured to load a global point cloud map of a location where the non-line-of-sight object is located, calibrate the perception device by using an automatic calibration algorithm, and obtain a global pose of the perception device in the global point cloud map of the location where the non-line-of-sight object is located.

[0026] A local pose calculation module is configured to identify the non-line-of-sight object entering the global point cloud map by using an automatic identification algorithm, and calculate a local pose of the non-line-of-sight object relative to the perception device.

[0027] A pose conversion module is configured to convert the local pose of the non-line-of-sight object into a global pose by using a coordinate conversion algorithm and combining the global pose of the perception device.

[0028] An identification module is configured to identify a target point by using the converted global pose as a target point.

[0029] In a third aspect, an apparatus for identifying a non-line-of-sight object is provided, which includes:

[0030] A perception device is arranged in a region covering the non-line-of-sight object, and is configured to collect point cloud data of the non-line-of-sight object.

[0031] A calibration module is connected to the perception device, and is configured to calibrate a global pose of the perception device by using a global point cloud map of a location where the non-line-of-sight object is located and by calling a point cloud global positioning matching method.

[0032] A calculation module is connected to the perception device and the calibration module, and is configured to identify the non-line-of-sight object and calculate a local pose of the non-line-of-sight object relative to the perception device by calling an automatic identification algorithm.

[0033] A conversion module is connected to the calculation module, and is configured to convert the local pose of the non-line-of-sight object into a global pose by using a rotation matrix inverse operation based on the global pose of the perception device.

[0034] A communication identification module is connected to the conversion module, and is configured to receive global pose information converted by the conversion module, and generate a scheduling command by using the received global pose information as a target point for an identification task.

[0035] In a fourth aspect, a device is provided, and the device is arranged with the apparatus for identifying a non-line-of-sight object.

[0036] The main advantages of the technical scheme of the present application are as follows:

[0037] The method, system, device and equipment for identifying non-line-of-sight objects provided by the application can provide high-precision environment perception through solid-state laser radar, enhance the robustness of target recognition by combining deep learning algorithm, quickly complete the registration of radar and global map based on the point cloud global positioning algorithm of iterative matching, realize real-time conversion of local and global coordinate systems through inverse operation of rotation matrix, ensure positioning accuracy, and improve the accuracy of non-line-of-sight scene cargo location point recognition to more than 95% in combination with actual application, shorten the calibration time to within 30 seconds, significantly improve the system deployment efficiency, support multi-AGV collaborative work, and adapt to complex logistics environment. BRIEF DESCRIPTION OF DRAWINGS

[0038] The drawings described herein are used to provide further understanding of the embodiments of the application, and form a part of the application. The illustrative embodiments of the application and their descriptions are used to explain the application, and do not constitute an improper limitation on the application. In the drawings:

[0039] Figure 1 A flowchart of a method for identifying non-line-of-sight objects provided by an embodiment of the application is shown in the figure;

[0040] Figure 2 A schematic diagram of a method for identifying non-line-of-sight objects provided by an embodiment of the application is shown in the figure;

[0041] Figure 3 A schematic diagram of a system for identifying non-line-of-sight objects provided by an embodiment of the application is shown in the figure;

[0042] Figure 4 A schematic diagram of a device for identifying non-line-of-sight objects provided by an embodiment of the application is shown in the figure;

[0043] Explanation of reference signs:

[0044] 10, observation module; 20, global pose calibration module; 30, local pose calculation module; 40, pose conversion module; 50, identification module;

[0045] 100, perception equipment; 200, calibration module; 300, calculation module; 400, conversion module; 500, communication identification module. DETAILED DESCRIPTION

[0046] To make the purpose, technical solutions and advantages of the application clearer, the technical solutions of the application will be described clearly and completely below in combination with specific embodiments of the application and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the application, not all the embodiments. Based on the embodiments of the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the application.

[0047] The technical solutions provided by the embodiments of the present application are described in detail below in combination with the accompanying drawings 1-4.

[0048] In order to more clearly illustrate the technical solutions of the present application, the "non-line-of-sight" involved in the present application is further explained and described.

[0049] Non-line-of-sight (NLOS) refers to a scenario in which a target object cannot be observed or perceived through a direct line of sight (such as vision, laser, etc.) due to the presence of physical obstacles. In the embodiments of the present application, non-line-of-sight refers to the existence of an obstruction (such as a cargo box, a wall, other goods, etc.) between a forklift or an automated guided vehicle (AGV) and a target storage location (such as a shelf, a pallet), which prevents the forklift's own vision system or sensor from directly observing the target location. The scenario applied in the embodiments of the present application is that the forklift is located outside the truck bed, and the target storage location inside the cargo box is blocked by the truck bed wall or other goods, and cannot be directly positioned by the camera or traditional sensor. The shelf or pallet is located in the corner of the warehouse, behind the multi-layer shelf, or is blocked by other handling equipment.

[0050] In order to solve the above-mentioned problem of coping with non-line-of-sight, the embodiments of the present application penetrate through part of the obstruction by three-dimensional point cloud, obtain the geometric information of the target object, pre-construct a high-precision environment map, assist in calibration and coordinate conversion, fuse point cloud and vision data, improve the robustness of target recognition in the occluded scene, match local perception data with the global map, and indirectly calculate the global pose of the target. Specifically, in the embodiments of the present application, the truck bed outside is scanned by a laser radar, the structure inside the truck bed is inferred in combination with the global map, the point cloud features of the shelf inside the truck bed are identified by a depth model, the relative position thereof is calculated, the local coordinates are mapped to the global map through coordinate conversion, the unloading target point is generated, and precise recognition and positioning are achieved.

[0051] The following will be described in combination with the accompanying drawings Figures 1-4 The present application is how to achieve precise recognition and positioning.

[0052] Embodiment one

[0053] As shown in the accompanying drawings Figures 1-2 The embodiments of the present application provide a method for identifying non-line-of-sight objects, which comprises the following steps S1-S5:

[0054] Step S1: setting a perception device, so that the set perception device can directly observe the non-line-of-sight object and the area where it is located;

[0055] In some optional implementations of the embodiment, the perception device is set as a perception sensor. For example, a solid-state laser radar is fixedly installed at a fixed position such as the top of a warehouse shelf or a wall surface according to actual application, and the field of view of the solid-state laser radar covers the pallet parking area where the non-line-of-sight object is located.

[0056] In some other optional implementations of the embodiment, the solid-state laser radar is also fixedly installed at a position where the pallet can be directly observed for collecting point cloud data of the pallet and the pallet according to actual application.

[0057] In order to ensure that the perception device can accurately observe the point cloud data of the non-line-of-sight object, the solid-state laser radar is set as an Ouster OS1-64, a Velodyne VLP-16 or a similar model of laser radar device.

[0058] As can be seen, in the embodiment of the application, by using the perception device and setting the perception device as a solid-state laser radar, the perception device also has the ability to provide high-precision environmental perception, which can guarantee accurate identification of the non-line-of-sight pallet position of the forklift, improve the identification accuracy, and improve the unloading efficiency.

[0059] It can be understood that the number of the set solid-state laser radars includes multiple, and the specific number is determined according to actual needs.

[0060] In some optional implementations of the embodiment, the solid-state laser radar is taken as a perception sensor, the shelf or pallet required by the forklift for unloading is taken as a non-line-of-sight object, and the method for accurately identifying the non-line-of-sight pallet position in the forklift unloading scene is taken as an example to illustrate the specific principle and implementation manner of the application.

[0061] Step S2: loading a global point cloud map of the position where the non-line-of-sight object is located, calibrating the perception device by using an automatic calibration algorithm, and obtaining a global pose of the perception device under the global point cloud map of the position where the non-line-of-sight object is located.

[0062] In some optional implementations of the embodiment, the automatic calibration algorithm includes a point cloud global positioning matching method.

[0063] Specifically, the automatic calibration algorithm is used to calibrate the perception device, and a global pose of the perception device under the global point cloud map of the position where the non-line-of-sight object is located is obtained.

[0064] The initial pose parameter of the perception device is input, the point cloud data collected by the perception device in real time is matched with the global point cloud map by using the automatic calibration algorithm, the initial pose parameter is iteratively optimized until the matching error is less than a preset threshold.

[0065] In combination with the actual application, in some optional implementation manners of the embodiment of the present application, the preset threshold value is 5 cm, and of course the preset threshold value can also be set to any value in the range of 3 cm-15 cm, such as 3 cm, 4 cm,..., 13 cm, 14 cm and 15 cm, and of course it can be understood that the above-mentioned preset threshold value is only illustratively described and is not the only limitation of the embodiment of the present application, and the specific preset threshold value can also be determined according to actual needs through calculation or prior knowledge.

[0066] In some optional implementation manners of the embodiment, the global point cloud map can be constructed in the following manner:

[0067] The mobile robot is used to carry a laser radar to perform three-dimensional scanning on the working area; a SLAM (simultaneous localization and mapping) algorithm is used to process the scanning data to generate a global point cloud map containing the positions of non-line-of-sight objects; and then the global point cloud map is stored in a server or a local storage of the perception device.

[0068] Step S3: An automatic recognition algorithm is used to recognize the non-line-of-sight objects entering the global point cloud map and calculate the local pose of the non-line-of-sight objects relative to the perception device.

[0069] Illustratively, in the embodiment of the present application, the automatic recognition algorithm includes a deep learning model, the training of the deep learning model and the input data include non-line-of-sight object images and point cloud data under multiple angles and multiple light conditions, and the output of the deep learning model is the local pose of the non-line-of-sight object.

[0070] In another optional implementation manner of the embodiment, in order to make the detection data acquired by the perception device more accurate and guarantee the training structure of the deep model, three-dimensional space coordinate information of the target shelf or tray can also be acquired to cover the point cloud distribution under different distances, angles and occlusion conditions, such as RGB or grayscale images acquired by a camera, containing the texture, color and edge features of the tray, enhancing the recognition ability of the model under complex light conditions.

[0071] Therefore, the present application can cover different warehouse layouts, such as narrow passages, multi-layer shelves, light conditions, such as natural light, artificial lighting, low light environment, weather influences, such as sensor noise caused by rainy days and foggy days, thereby meeting the characteristics of scene diversity, and in addition, it can also include various tray types, such as wood, plastic and metal; shelf forms, such as single-layer or multi-layer and possible occlusion conditions, such as goods stacking or forklift arm occlusion, etc., realizing the characteristics of target diversity.

[0072] Therefore, in combination with actual application, by collecting tray images and point cloud data under different angles and light conditions, target detection is performed by using a YOLOv5 model, and local poses (X, Y, θ) of the tray are output in combination with point cloud feature extraction, so that accurate positioning can be realized, and inaccurate recognition caused by insufficient or excessive light and further caused recognition errors and affecting the working efficiency and quality of the forklift are avoided.

[0073] In some optional implementations of the embodiment, the automatic recognition algorithm is implemented based on a deep learning model, and examples are as follows:

[0074] For example, point cloud data collected by the perception device is processed by using a point cloud deep learning model such as PointNet++ or PointRCNN, or image data collected by the perception device is processed by using a visual recognition model such as YOLO or Faster R-CNN, the type, position and pose of the non-line-of-sight object are recognized by the model, and the coordinates (u, v, w) and the pose angle (α, β, γ) of the center point in the local coordinate system are output; it should be noted that the above algorithm is a prior art algorithm, and in the embodiment of the application, the data obtained by the embodiment of the application is only used for training and application, and since it is a prior art algorithm, it is only used in the process, and thus it is clear and can be implemented in the embodiment of the application, and thus it is not described in detail in the embodiment of the application.

[0075] Therefore, in the embodiment of the application, multiple trays or shelves can be recognized simultaneously by using a deep learning model and an automatic recognition algorithm, and the pose information of each tray or shelf is output, the purpose of multi-target recognition is achieved, the model can be continuously updated through online learning to adapt to changes in warehouse layout or newly added shelf types, the purpose of dynamic environment adaptation is achieved, in addition, the model of the embodiment of the application can be adapted to different brands of laser radars (such as Velodyne and Livox) and AGV scheduling systems (such as ROS and MiR), and thus the deep learning model of the application realizes high-precision storage location recognition in a non-line-of-sight scene by fusing laser radar point cloud and visual data, combining automatic calibration and coordinate conversion technology. The diversity of training data, the optimization of model architecture and the strict verification process ensure the reliability and real-time performance of the system in a complex logistics environment.

[0076] Step S4: converting the local pose of the non-line-of-sight object into a global pose by a coordinate conversion algorithm in combination with the global pose calibrated by the perception device;

[0077] Specifically, the local pose of the non-line-of-sight object is converted into a global pose by a coordinate conversion algorithm in combination with the global pose of the perception device, and the conversion is determined by inverse rotation matrix calculation, wherein the inverse rotation matrix calculation includes:

[0078]

[0079] wherein, represents the local pose of the non-line-of-sight object relative to the perception device, represents the global pose of the perception device, represents the local pose of the non-line-of-sight object.

[0080] In some other optional implementations of the embodiments of the present application, the coordinate conversion algorithm can also use homogeneous transformation matrix operations, which include, for example:

[0081]

[0082] wherein, represents the global pose of the non-line-of-sight object, represents the global pose of the perception device as a homogeneous transformation matrix, represents the local pose of the non-line-of-sight object.

[0083] Thus, the pose error of the solid-state laser radar can be eliminated through inverse matrix operations and homogeneous transformation matrix operations to ensure conversion accuracy.

[0084] Step S5: The converted global pose is taken as a target point position, and the identification task is performed.

[0085] In combination with actual applications, for example, the information of this target point position is synchronized to the AGV scheduling system in real time, and the scheduling system takes it as a target point position to issue a scheduling command to the AGV forklift, thereby realizing the precise unloading task of the AGV forklift.

[0086] In summary, a method for accurately identifying a non-line-of-sight cargo point position in a forklift unloading scenario using the method for identifying a non-line-of-sight object according to the embodiments of the present application can realize the identification of a non-line-of-sight object and the calculation of a local pose, and in combination with a coordinate conversion algorithm, the local pose can be further converted into a target point position in a global coordinate system to provide accurate navigation for a forklift AGV. Specifically, when unloading cargo from a truck, the method for accurately identifying non-line-of-sight cargo position information includes the following processes and principles:

[0087] A solid-state lidar is used as a perception sensor, installed in a fixed position where the cargo box can be directly observed. Once the solid-state lidar is installed, the global point cloud map is automatically loaded, and the automatic calibration algorithm is activated to complete radar calibration, obtaining the radar's global pose within the global map. When a cargo vehicle enters the system's line of sight, the shelf recognition algorithm is activated to identify the shelf (or pallet) and obtain its local pose relative to the radar. Finally, a coordinate conversion algorithm is used, combined with the radar's own global pose, to convert the local pose of the shelf (or pallet) into a global pose. This information is synchronized in real time to the AGV dispatching system, which uses it as the target point and issues dispatch commands to the AGV forklift. After receiving the global pose, the dispatching system plans the optimal path; the AGV forklift follows the instructions and drives to the target point, with an error control within ±2 cm. This enables the AGV forklift to accurately unload cargo.

[0088] Therefore, a method for identifying non-line-of-sight objects in an embodiment of the present invention provides high-precision environmental perception through solid-state laser radar, combines deep learning algorithms to enhance target recognition robustness, and uses a point cloud global positioning algorithm based on iterative matching to quickly complete the alignment of the radar and the global map. It realizes real-time conversion of local and global coordinate systems through inverse rotation matrix operations to ensure positioning accuracy. Combined with practical applications, it can improve the accuracy of cargo location identification in non-line-of-sight scenarios to more than 95%, shorten the calibration time to within 30 seconds, significantly improve system deployment efficiency, support multi-AGV collaborative operation, and adapt to complex logistics environments.

[0089] Example 2

[0090] The embodiment of the present invention also provides a system for identifying non-line-of-sight objects. Figure 3 As shown, it includes: an observation module 10, a global pose calibration module 20, a local pose calculation module 30, a pose conversion module 40 and a recognition module 50, wherein:

[0091] The observation module 10 is used to set up a sensing device so that the sensing device can directly observe the non-line-of-sight object and the area where it is located; the global pose calibration module 20 is used to load the global point cloud map of the location of the non-line-of-sight object, and calibrate the sensing device using an automatic calibration algorithm to obtain the global pose of the sensing device under the global point cloud map of the location of the non-line-of-sight object; the local pose calculation module 30 is used to use an automatic recognition algorithm to identify the non-line-of-sight object entering the global point cloud map, and calculate the local pose of the non-line-of-sight object compared to the sensing device; the pose conversion module 40 is used to convert the local pose of the non-line-of-sight object into a global pose through a coordinate conversion algorithm combined with the global pose obtained by the calibration of the sensing device; the recognition module 50 is used to use the converted global pose as the target point to perform the recognition task.

[0092] In an embodiment of the present invention, a system for identifying non-line-of-sight objects is independent of the forklift body and can be seamlessly connected to the existing AGV scheduling system, thereby automating the entire process from identification, calibration to coordinate conversion, thereby improving unloading efficiency.

[0093] Example 3

[0094] The embodiment of the present invention also provides a device for identifying non-line-of-sight objects, such as Figure 4 As shown, it includes: a sensing device 100, a calibration module 200, a calculation module 300, a conversion module 400 and a communication identification module 500, wherein:

[0095] The perception device 100 is set in an area covering the location of non-line-of-sight objects and is used to collect point cloud data of non-line-of-sight objects; the calibration module 200 is connected to the perception device 100, and the calibration module 200 is based on the global point cloud map of the location of the non-line-of-sight objects. The calibration module 200 completes the global pose calibration of the perception device 100 by calling the point cloud global positioning matching method; the calculation module 300 is connected to the perception device 100 and the calibration module 200, and the calculation module 300 is used to identify the non-line-of-sight objects and calculate their local pose relative to the perception device 100 by calling the automatic recognition algorithm; the conversion module 400 is connected to the calculation module 300, and the conversion module 400 is used to convert the local pose of the non-line-of-sight objects into a global pose based on the global pose of the perception device 100 through the inverse operation of the rotation matrix; the communication identification module 500 is connected to the conversion module 400, and is used to receive the global pose information converted by the conversion module 400, and use the received global pose information as the target point to generate a scheduling command and output it to the execution unit to perform the recognition task.

[0096] In an embodiment of the present invention, a device for identifying non-line-of-sight objects provided by the embodiment of the present invention is independent of the forklift body. Since the laser radar is not affected by light or occlusion, it is suitable for complex logistics environments and has strong non-line-of-sight adaptability. Through global point cloud matching and coordinate transformation, it can achieve millimeter-level positioning accuracy, and thus has high-precision positioning capabilities. Combined with practical applications, since the device is independent of the forklift body, it can be seamlessly connected to the existing AGV scheduling system, so the compatibility of the present invention is good. Therefore, the device of the present invention can automate the entire process from identification, calibration to coordinate conversion, thereby improving unloading efficiency. It can seamlessly connect to the existing AGV scheduling system, thereby enabling the device of the present invention to automate the entire process from identification, calibration to coordinate conversion, thereby improving unloading efficiency.

[0097] Therefore, the device provided by the embodiment of the present application can provide high-precision environment perception through a solid-state laser radar, enhance the robustness of target recognition by combining a deep learning algorithm, quickly complete the registration of the radar and a global map based on an iterative matching point cloud global positioning algorithm, realize real-time conversion between a local coordinate system and a global coordinate system through inverse rotation matrix operation, ensure positioning accuracy, improve the accuracy of the identification of the storage location in a non-line-of-sight scene to more than 95%, shorten the calibration time to within 30 seconds, significantly improve the system deployment efficiency, support collaborative work of multiple AGVs, and adapt to complex logistics environments.

[0098] Embodiment four

[0099] The embodiment of the present application also provides a device, which is provided with the device for identifying non-line-of-sight objects.

[0100] Through the device of the present application, high-precision environment perception can be provided through a solid-state laser radar, the robustness of target recognition can be enhanced by combining a deep learning algorithm, the registration of the radar and a global map can be quickly completed based on an iterative matching point cloud global positioning algorithm, real-time conversion between a local coordinate system and a global coordinate system can be realized through inverse rotation matrix operation, positioning accuracy can be ensured, the accuracy of the identification of the storage location in a non-line-of-sight scene can be improved to more than 95%, the calibration time can be shortened to within 30 seconds, the system deployment efficiency can be significantly improved, collaborative work of multiple AGVs can be supported, and complex logistics environments can be adapted.

[0101] It should be noted that, in this document, relational terms such as "first" and "second", and the like, are used solely to distinguish one entity or action from another entity or action, without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. Furthermore, the terms "front", "rear", "left", "right", "upper", "lower", and the like, are used only to denote the positions of the components shown in the drawings, and are not intended to limit the scope of the present application.

[0102] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for identifying non-line-of-sight objects, characterized in that: include: Setting up sensing equipment so that the sensing equipment can directly observe non-line-of-sight objects and the areas where they are located; Load the global point cloud map of the location of the non-line-of-sight object, calibrate the perception device using the automatic calibration algorithm, and obtain the global pose of the perception device under the global point cloud map of the location of the non-line-of-sight object; Use an automatic recognition algorithm to identify non-line-of-sight objects that enter the global point cloud and calculate the local pose of the non-line-of-sight objects relative to the sensing device; The local pose of the non-line-of-sight object is converted into a global pose through a coordinate conversion algorithm combined with the global pose obtained by the calibration of the sensing device. The rotated global pose is used as the target point to perform the recognition task.

2. The method for identifying non-line-of-sight objects according to claim 1, wherein: The sensing device includes a sensing sensor.

3. The method for identifying non-line-of-sight objects according to claim 1, wherein: The automatic calibration algorithm is used to calibrate the perception device, and the global pose of the perception device under the global point cloud map at the location of the non-line-of-sight object is obtained, including: Input the initial pose parameters of the perception device; Match the point cloud data collected by the sensing device in real time with the global point cloud map through an automatic calibration algorithm; The initial pose parameters are iteratively optimized until the matching error is less than the preset threshold.

4. The method for identifying non-line-of-sight objects according to claim 3, wherein: The preset threshold includes 5 cm.

5. The method for identifying non-line-of-sight objects according to claim 3, wherein: The automatic calibration algorithm includes: point cloud global positioning matching method.

6. The method for identifying non-line-of-sight objects according to claim 1, wherein: The automatic recognition algorithm includes: a deep learning model. The training and input data of the deep learning model include non-line-of-sight object images and point cloud data under multiple angles and multiple lighting conditions. The output of the deep learning model is the local pose of the non-line-of-sight object.

7. The method for identifying non-line-of-sight objects according to claim 1, wherein: Through the coordinate transformation algorithm, combined with the global pose of the sensing device, the local pose of the non-line-of-sight object is converted to the global pose and determined by the inverse rotation matrix operation, where the inverse rotation matrix operation includes: in, Represents the local pose of the non-line-of-sight object relative to the sensing device, represents the global pose of the sensing device, Represents the local pose of non-line-of-sight objects.

8. A system for identifying non-line-of-sight objects, characterized in that: include: An observation module, wherein the observation module is configured to enable the configured sensing device to directly observe non-line-of-sight objects and the area in which they are located; A global pose calibration module is used to load a global point cloud map of the location of the non-line-of-sight object, calibrate the sensing device using an automatic calibration algorithm, and obtain the global pose of the sensing device under the global point cloud map of the location of the non-line-of-sight object; A local pose calculation module, which is used to use an automatic recognition algorithm to identify non-line-of-sight objects that enter the global point cloud point map and calculate the local pose of the non-line-of-sight objects compared to the sensing device; A posture conversion module is used to convert the local posture of the non-line-of-sight object into a global posture through a coordinate conversion algorithm combined with the global posture obtained by calibration of the sensing device; The recognition module is used to use the rotated global posture as the target point to perform the recognition task.

9. A device for identifying non-line-of-sight objects, characterized in that: include: A sensing device is provided in an area covering a non-line-of-sight object and is used to collect point cloud data of the non-line-of-sight object; A calibration module, connected to the sensing device, the calibration module is based on a global point cloud map of the location of non-line-of-sight objects, and the calibration module completes the global pose calibration of the sensing device by calling a point cloud global positioning matching method; a computing module connected to the sensing device and the calibration module, and configured to identify non-line-of-sight objects and calculate their local pose relative to the sensing device by invoking an automatic recognition algorithm; A conversion module, connected to the calculation module, configured to convert a local pose of a non-line-of-sight object into a global pose based on the global pose of the sensing device by performing an inverse rotation matrix operation; A communication identification module is connected to the conversion module, and is used to receive the global posture information converted by the conversion module, and use the received global posture information as the target point to generate a scheduling command to identify the operation task.

10. A device, provided with the apparatus for identifying non-line-of-sight objects according to claim 9.

Images