Map construction method, device and vehicle

Abstract

The present invention discloses a map construction method, device and vehicle. Among them, the method includes: obtaining multiple image frames collected by a target vehicle at different times, where the image frames are used to represent the environment around the target vehicle; determining the target predicted poses of the target vehicle at different times; based on the multiple image frames and the target predicted poses, determining the target registration scores of each image frame with a map frame, where the map frame is used to represent the image frame corresponding to a preset map model obtained by the target vehicle at different times; based on the target registration scores, determining target key frames among the multiple image frames; and processing the target key frames using the preset map model to obtain a target map. The present invention solves the technical problem in the prior art that due to inaccurate positioning of the vehicle in the underground garage, the accuracy of the constructed map is low.

Description

Technical Field

[0001] The present invention relates to the field of image recognition, and in particular, to a map construction method, device, and vehicle. Background Art

[0002] During the driving of a driverless vehicle, an accurate positioning function is an important condition to ensure the safety and stability of the driverless vehicle. Whether the positioning of the driverless vehicle is accurate depends to a large extent on the accuracy of the map of the area where it is located. The higher the accuracy, the more accurate the positioning of the driverless vehicle. However, in some areas with poor satellite signals, traditional map construction methods cannot well construct the real environmental conditions of these areas, resulting in a low accuracy of the constructed map.

[0003] For the above problems, no effective solution has been proposed yet. Summary of the Invention

[0004] Embodiments of the present invention provide a map construction method, device, and vehicle to at least solve the technical problem in the prior art that due to inaccurate positioning of the vehicle in the underground garage, the accuracy of the constructed map is low.

[0005] According to one aspect of the embodiments of the present invention, a map construction method is provided, including: obtaining multiple image frames collected by a target vehicle at different times, where the image frames are used to represent the environment around the target vehicle; determining the target predicted poses of the target vehicle at different times; based on the multiple image frames and the target predicted poses, determining the target registration scores of each image frame and a map frame, where the map frame is used to represent the image frame corresponding to a preset map model obtained by the target vehicle at different times; based on the target registration scores, determining the target key frames in the multiple image frames; and using the preset map model to process the target key frames to obtain a target map.

[0006] Optionally, determining target key frames in multiple image frames based on a target registration score includes: performing semantic segmentation on each image frame to obtain segmentation elements of each image, where the segmentation elements are used to characterize the recognition objects included in each image frame; obtaining a target speed and inertial measurement values of the target vehicle, where the target speed is used to characterize the driving speed of the target vehicle at different times, and the inertial measurement values are used to characterize the angular rate and acceleration of the target vehicle at different times; in response to the target speed and inertial measurement values being less than a first preset threshold, determining that the target vehicle is stationary, and storing the target image frame corresponding to the stationary target vehicle into a post-processing frame set, where the post-processing frame set is a set composed of image frames continuously obtained when the target vehicle is stationary; determining a first comprehensive score of each target image frame in the post-processing frame set based on the number of different types to which the segmentation elements corresponding to any target image frame in the post-processing frame set belong, first point cloud data, and the target registration score; in response to the first comprehensive score satisfying a first preset condition, determining the target image frame corresponding to the first comprehensive score as the target key frame.

[0007] Optionally, determining the target registration score of each image frame with respect to a map frame based on multiple image frames and a target predicted pose includes: obtaining second point cloud data of each image frame and third point cloud data of the map frame; determining the target registration score of each image frame with respect to the map frame by using a first preset algorithm based on the second point cloud data, the third point cloud data, and the target predicted pose.

[0008] Optionally, determining the target registration score of each image frame with respect to a map frame by using a first preset algorithm based on the second point cloud data, the third point cloud data, and the target predicted pose includes: updating the third point cloud data based on the target predicted pose to obtain fourth point cloud data; dividing the fourth point cloud data into multiple first divided point cloud data by using a preset conversion algorithm; converting the second point cloud data into second divided point cloud data corresponding to the first divided point cloud data by using a preset conversion matrix; and determining the target registration score by performing point cloud registration on the first divided point cloud data and the second divided point cloud data by using the first preset algorithm.

[0009] Optionally, obtaining multiple image frames collected by the target vehicle at different times includes: obtaining multiple target images corresponding to different times based on a preset time interval, where the preset time interval is used to characterize the time difference between different times, and the multiple target images are respectively used to characterize images obtained by multiple sensors based on which the target vehicle; and stitching the multiple target images by using a second preset algorithm to obtain the image frames collected at different times.

[0010] Optionally, determining the target predicted poses of the target vehicle at different times includes: in response to the presence of a target object in the image frame, determining a first calibration result of the target object, where the first calibration result is used to represent the coordinate position of the target object in a preset map model, and the target object is used to represent an object that affects the quality of the stitched image; obtaining a second calibration result of the target vehicle, where the second calibration result is used to represent the coordinate position of the target vehicle in the preset map model; in response to the second calibration result being different from the first calibration result, determining that the target vehicle is normal at the current time, and determining the target predicted pose at the current time based on a pose transformation algorithm and the second actual pose of the previous frame, where the previous frame is used to represent an image frame adjacent to and obtained before the current frame; in response to the second calibration result being the same as the first calibration result, determining that the target vehicle is abnormal at the current time and not determining the target predicted pose at the current time.

[0011] Optionally, after determining the target key frames in multiple image frames based on the target registration score, the method further includes: determining a first actual pose of the target key frame based on the inertial measurement value, the first point cloud data, and the target predicted pose by using a preset coupling algorithm; constructing a semantic matching factor based on the first actual pose and the target registration score, where the semantic matching factor is used to represent a factor for updating the target map; determining whether the target key frame and other key frames form a detection loop, where the detection loop is used to represent a dependency relationship between the target key frame and other key frames; in response to the target key frame and other key frames forming a detection loop, constructing a loop detection factor based on the target key frame and the map frame; updating the target map based on the semantic matching factor and the loop detection factor.

[0012] Optionally, determining whether the target key frame and other key frames form a detection loop includes: obtaining a change value between the third actual pose of the target key frame and the fourth actual pose of other key frames; determining whether the change value is less than a second preset threshold; in response to the change value being less than the second preset threshold, determining whether the time interval between the target key frame and other key frames is greater than a third preset threshold; in response to the time interval being greater than the third preset threshold, determining whether the difference between the second comprehensive score of the target key frame and the third comprehensive score of other key frames is less than a fourth preset threshold; in response to the difference being less than the fourth preset threshold, determining that the target key frame and other key frames form a detection loop.

[0013] According to another aspect of the embodiments of the present invention, there is also provided a map construction device, including: a first acquisition module, configured to acquire multiple image frames collected by a target vehicle at different times, where the image frames are used to characterize the environment around the target vehicle; a first determination module, configured to determine the target predicted poses of the target vehicle at different times; a second determination module, configured to determine the target registration scores of each image frame and a map frame based on the multiple image frames and the target predicted poses, where the map frame is used to characterize the image frame corresponding to a preset map model acquired by the target vehicle at different times; a third determination module, configured to determine target key frames among the multiple image frames based on the target registration scores; and a map update module, configured to process the target key frames using the preset map model to obtain a target map.

[0014] Optionally, the third determination module includes: a segmentation element determination unit, configured to perform semantic segmentation on each image frame to obtain the segmentation elements of each image, where the segmentation elements are used to characterize the recognition objects included in each image frame; a first acquisition unit, configured to acquire the target speed and inertial measurement values of the target vehicle, where the target speed is used to characterize the driving speed of the target vehicle at different times, and the inertial measurement values are used to characterize the angular rate and acceleration of the target vehicle at different times; a first storage unit, configured to determine that the target vehicle is stationary in response to the target speed and inertial measurement values being less than a first preset threshold, and store the target image frame corresponding to the stationary state of the target vehicle into a post-processing frame set, where the post-processing frame set is a set composed of image frames continuously acquired when the target vehicle is stationary; a comprehensive score determination unit, configured to determine the first comprehensive score of each target image frame in the post-processing frame set based on the number of different types to which the segmentation elements corresponding to any target image frame in the post-processing frame set belong, the first point cloud data, and the target registration score; and a target key frame determination unit, configured to determine the target image frame corresponding to the first comprehensive score as the target key frame in response to the first comprehensive score satisfying a first preset condition.

[0015] Optionally, the second determination unit includes: a second acquisition unit, configured to acquire the second point cloud data of each image frame and the third point cloud data of the map frame; and a registration score determination unit, configured to determine the target registration score of each image frame and the map frame based on the second point cloud data, the third point cloud data, and the target predicted pose using a first preset algorithm.

[0016] Optionally, the registration score determination unit includes: a first determination subunit, configured to update the third point cloud data based on the target predicted pose to obtain fourth point cloud data; a first division subunit, configured to divide the fourth point cloud data into multiple first divided point cloud data by using a preset conversion algorithm; a first conversion subunit, configured to convert the second point cloud data into second divided point cloud data corresponding to the first divided point cloud data by using a preset conversion matrix; and a first registration subunit, configured to perform point cloud registration on the first divided point cloud data and the second divided point cloud data by using a first preset algorithm to determine a target registration score.

[0017] Optionally, the first acquisition module includes: a target image acquisition unit, configured to acquire multiple target images corresponding to different moments based on a preset time interval, where the preset time interval is used to represent the time difference between different moments, and the multiple target images are respectively used to represent images acquired by multiple sensors based on a target vehicle; and an image stitching unit, configured to stitch the multiple target images based on a second preset algorithm to obtain an image frame acquired at different moments.

[0018] Optionally, the first determination module includes: a first calibration determination unit, configured to determine a first calibration result of a target object in response to the presence of the target object in the image frame, where the first calibration result is used to represent the coordinate position of the target object in a preset map model, and the target object is used to represent an object affecting the quality of the stitched image; a second calibration acquisition unit, configured to acquire a second calibration result of the target vehicle, where the second calibration result is used to represent the coordinate position of the target vehicle in the preset map model; and a target predicted pose determination unit, configured to determine that the target vehicle is normal at the current moment in response to the second calibration result being different from the first calibration result, and determine the target predicted pose at the current moment based on a pose transformation algorithm and the second actual pose of the previous frame, where the previous frame is used to represent an image frame adjacent to and acquired before the current frame; and determine that the target vehicle is abnormal at the current moment and not determine the target predicted pose at the current moment in response to the second calibration result being the same as the first calibration result.

[0019] Optionally, the device further includes: an actual pose determination module, configured to determine a first actual pose of a target key frame based on inertial measurement values, first point cloud data, and a target predicted pose by using a preset coupling algorithm; a matching factor construction module, configured to construct a semantic matching factor based on the first actual pose and a target registration score, where the semantic matching factor is used to represent a factor for updating a target map; a loop detection and judgment module, configured to judge whether the target key frame and other key frames form a detection loop, where the detection loop is used to represent a dependency relationship between the target key frame and other key frames; a detection factor construction module, configured to, in response to the target key frame and other key frames forming a detection loop, construct a loop detection factor based on the target key frame and a map frame; and a target map update module, configured to update the target map based on the semantic matching factor and the loop detection factor.

[0020] Optionally, the loop detection and judgment module includes: a third acquisition unit, configured to acquire a change value between a third actual pose of the target key frame and a fourth actual pose of other key frames; a first judgment unit, configured to judge whether the change value is less than a second preset threshold; a second judgment unit, configured to, in response to the change value being less than the second preset threshold, judge whether a time interval between the target key frame and other key frames is greater than a third preset threshold; a third judgment unit, configured to, in response to the time interval being greater than the third preset threshold, judge whether a difference between a second comprehensive score of the target key frame and a third comprehensive score of other key frames is less than a fourth preset threshold; and a loop determination unit, configured to, in response to the difference being less than the fourth preset threshold, determine that the target key frame and other key frames form a detection loop.

[0021] On the other hand, according to an embodiment of the present invention, a computer-readable storage medium is further provided, including a stored program, where, when the program runs, it controls a device where the computer-readable storage medium is located to execute the map construction method according to any one of the above.

[0022] On the other hand, according to an embodiment of the present invention, a processor is further provided, where, when a program of the processor runs, it executes the map construction method according to any one of the above.

[0023] On the other hand, according to an embodiment of the present invention, a target vehicle is further provided, including: one or more processors; a storage device, configured to store one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors execute the map construction method according to any one of the above.

[0024] In the embodiments of the present invention, multiple image frames collected by a target vehicle at different times are acquired; target predicted poses of the target vehicle at different times are determined; based on the multiple image frames and the target predicted poses, target registration scores of each image frame with a map frame are determined; based on the target registration scores, target key frames among the multiple image frames are determined; and a preset map model is used to process the target key frames to obtain a target map. In this way, the surrounding environment of the target vehicle is directly recognized and detected, improving the robustness to scene changes during map construction. At the same time, by calculating the target registration scores of each image frame, target key frames with higher quality are determined, and the target key frames are used to construct and optimize a preset initial map model, greatly improving the quality of the constructed target map, and solving the technical problem in the prior art that the accuracy of the constructed map is low due to inaccurate positioning of the vehicle in an underground garage. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The drawings described herein are used to provide a further understanding of the present invention, and constitute a part of this application. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention. In the drawings:

[0026] Figure 1 is a flowchart of the steps of a map construction method shown according to an embodiment of the present invention;

[0027] Figure 2 is a schematic diagram of the modules of a map construction system shown according to an embodiment of the present invention;

[0028] Figure 3 is a flowchart of the steps of a method for determining target key frames shown according to an embodiment of the present invention;

[0029] Figure 4 is a schematic flowchart of semantic feature preprocessing shown according to an embodiment of the present invention;

[0030] Figure 5 is a schematic flowchart of local map construction shown according to an embodiment of the present invention;

[0031] Figure 6 is a block diagram of the structure of a map construction device shown according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0032] To enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0033] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described herein can be implemented in an order different from those illustrated or described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily need to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0034] Currently, common map construction methods are generally divided into two categories: one is to perform point cloud recognition on images to construct a map. This method directly performs point cloud recognition on the environment around the vehicle, making the constructed map have high accuracy. However, each recognition is accompanied by a large amount of point cloud data, which is difficult to store and is easily affected by environmental changes; the other is to perform semantic segmentation detection on images to reduce the detection interference caused by environmental changes and light. However, the map constructed by this method is largely affected by the detection results. If misdetection or undetected detection occurs during detection, it is very easy to cause the finally constructed map to be incomplete or inaccurate.

[0035] Embodiment 1

[0036] In order to improve the accuracy of the map constructed for areas with poor communication quality, according to the embodiments of the present invention, a method embodiment for map construction is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. And although the logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in a different order from here.

[0037] Figure 1 is a step flowchart of a map construction method shown according to the embodiments of the present invention. As Figure 1 shown, the method includes the following steps:

[0038] Step S102: Obtain multiple image frames collected by the target vehicle at different times.

[0039] Among them, the image frames are used to represent the environment around the target vehicle.

[0040] Generally, when constructing a target map for areas with poor communication signal quality, such as underground garages and tunnels, where information cannot even be received, it is often necessary to use surveying equipment to conduct regional autonomous target map construction. Using a vehicle for target map construction is one of the most common methods. The above-mentioned target vehicle can refer to a vehicle used for target map construction. The above-mentioned image frames can refer to the environmental images obtained at different times by using a camera device, such as a camera, when the vehicle is driving in areas with poor communication signals, such as underground garages and tunnels.

[0041] Optionally, an image frame can be collected every once in a while, for example, once every 0.5 s, to ensure that the system will not be overloaded due to too many image frames to be processed during target map construction.

[0042] Step S104: Determine the target predicted poses of the target vehicle at different times.

[0043] The above-mentioned target predicted poses can refer to the poses of the surveying equipment. Here, it refers to the poses of the target vehicle used for target map construction. During the process of target map construction, the above-mentioned target predicted poses can refer to the transformation matrix that transforms the position of the target vehicle in the world coordinate system of the above-mentioned area into the position in the camera coordinate system of the above-mentioned camera device, which can be obtained through mapping, translation, or rotation. Specific techniques can refer to relevant literature and will not be elaborated here.

[0044] After collecting an image frame, the position of the target vehicle can be determined based on this image frame, and then the target predicted pose of the target vehicle can be determined.

[0045] Step S106: Based on multiple image frames and target predicted poses, determine the target registration scores between each image frame and the map frame.

[0046] Among them, the map frame is used to represent the image frame corresponding to the preset target map model obtained by the target vehicle at different times.

[0047] Generally, when conducting target map construction, an initial map model can be preset in the target map construction system. When the target vehicle collects multiple image frames, the system can use the collected images to shape the initial map model, and finally form a complete actual target map. The above-mentioned map frame can refer to the image frame that can be collected by the target vehicle simulated according to the position of the target vehicle in the initial map model.

[0048] The above target registration score can be used to represent the quality of the acquired image frames. The higher the target registration score, the higher the quality of the corresponding image frame.

[0049] Step S108: Based on the target registration score, determine the target key frames among multiple image frames.

[0050] The above target key frames can refer to the image frames used for target map construction. After multiple image frames are acquired, the target registration score of each image frame can be used to determine the image frame with higher quality as the target key frame to construct the target map, thereby improving the accuracy of the constructed target map.

[0051] Step S110: Process the target key frames using a preset map model to obtain the target map.

[0052] After the target key frames are determined, the information of these target key frames can be mapped into a preset initial map model to shape the initial map model and determine the final target map.

[0053] In the embodiment of the present invention, a method is adopted, which includes acquiring multiple image frames collected by a target vehicle at different times; determining the target predicted poses of the target vehicle at different times; based on the multiple image frames and the target predicted poses, determining the target registration scores of each image frame and a map frame; based on the target registration scores, determining the target key frames among the multiple image frames; and processing the target key frames using a preset map model to obtain the target map. This method directly performs recognition and detection on the surrounding environment of the target vehicle, improves the robustness to scene changes during map construction, and at the same time determines the target key frames with higher quality by calculating the target registration scores of each image frame, and uses the target key frames to optimize the construction of the preset initial map model, greatly improving the quality of the constructed target map, and solving the technical problem in the prior art that the accuracy of the constructed map is low due to inaccurate positioning of the vehicle in the underground garage.

[0054] Figure 2 It is a schematic diagram of the modules of a map construction system shown according to an embodiment of the present invention, as Figure 2As shown, the system may include five modules. First is the sensor information preprocessing module, which can preprocess the environmental information used to construct the target map. Second is the semantic feature preprocessing module, which can perform secondary preprocessing on the preprocessed environmental information to further improve the quality of the obtained environmental information. Then is the local map construction module, which can preliminarily construct the target map according to the environmentally processed information after the secondary preprocessing. Next is the loop detection module, which can determine whether the constructed target map can be optimized according to the environmentally processed information after the secondary preprocessing. Finally is the joint optimization module, which can extract some environmental factors of the environmental information for constructing the target map to optimize and update the constructed target map. The specific functions of each module can be referred to in the following text.

[0055] Optionally, determining the target key frames in multiple image frames based on the target registration score includes: performing semantic segmentation on each image frame to obtain the segmentation elements of each image, where the segmentation elements are used to represent the recognition objects included in each image frame; obtaining the target speed and inertial measurement values of the target vehicle, where the target speed is used to represent the driving speed of the target vehicle at different times, and the inertial measurement values are used to represent the angular rate and acceleration of the target vehicle at different times; in response to the target speed and inertial measurement values being less than the first preset threshold, determining that the target vehicle is stationary, and storing the target image frame corresponding to the stationary target vehicle in the post-processing frame set, where the post-processing frame set is a set composed of continuously acquired image frames when the target vehicle is stationary; based on the number of different types to which the segmentation elements corresponding to any target image frame in the post-processing frame set belong, the first point cloud data, and the target registration score, determining the first comprehensive score of each target image frame in the post-processing frame set; in response to the first comprehensive score satisfying the first preset condition, determining the target image frame corresponding to the first comprehensive score as the target key frame.

[0056] The method for determining the target key frames from the collected multiple image frames is as Figure 3 shown Figure 3 is a flowchart of the steps of a method for determining target key frames according to an embodiment of the present invention. The specific steps are as follows:

[0057] Step S302, perform semantic segmentation on each image frame to obtain the segmentation elements of each image.

[0058] Among them, the segmentation elements are used to represent the recognition objects included in each image frame.

[0059] In underground parking scenarios, lane lines, parking spaces, road markings and other element features can generally be used to determine the location of the target vehicle. Therefore, after acquiring an image frame, in order to improve the accuracy of the constructed target map, the image frame can be semantically segmented to identify these element features. Based on these element features, the current position of the target vehicle can be determined to avoid inaccurate positioning of the target vehicle due to poor communication quality.

[0060] Step S304: Acquire the target speed and inertia measurement value of the target vehicle.

[0061] The target speed is used to characterize the driving speed of the target vehicle at different times, and the inertial measurement value is used to characterize the angular rate and acceleration of the target vehicle at different times.

[0062] During the image frame acquisition process, the IMU (Inertial Measurement Unit) on the target vehicle can be used to obtain the current driving speed and inertial measurement values in real time, and determine whether the target vehicle is currently in motion based on the driving speed and inertial measurement values. Specifically, a first preset threshold of 0 can be set. If the current driving speed and inertial measurement values of the target vehicle are both 0, it can be determined that the target vehicle is currently not moving.

[0063] In an optional solution of this embodiment, different thresholds can be set for the above-mentioned driving speed and inertia measurement values to determine whether the target vehicle is currently in motion according to actual conditions. For example, the corresponding threshold for driving speed is set to 0.2m / s, the corresponding threshold for angular velocity is set to 0.02rad / s, and the corresponding threshold for acceleration is set to 0.03m / s. 2 wait.

[0064] Step S306 : In response to the target speed and the inertia measurement value being less than the first preset threshold, it is determined that the target vehicle is stationary, and a target image frame corresponding to the stationary target vehicle is stored in a post-processing frame set.

[0065] The post-processing frame set is a set of image frames continuously acquired when the target vehicle is stationary.

[0066] If the target vehicle is stationary, that is, not in motion, the environment around the vehicle will not change. At this time, there will be a lot of repeated environmental information in the multiple image frames collected. Considering the computational pressure of the system when building the target map, and in order to improve the robustness and accuracy of the constructed target map, the image frames continuously collected when the target vehicle is stationary can be stored in a post-processing set, and then the image frame with the highest quality is determined from these continuously collected image frames as the target key frame to build the target map.

[0067] Step S308: Determine the first comprehensive score of each target image frame in the post - processing frame set based on the number of different types to which the segmentation elements corresponding to any target image frame in the post - processing frame set belong, the first point cloud data, and the target registration score.

[0068] Specifically, the first comprehensive score of each of the above - mentioned multiple consecutive image frames can be determined according to the number of types to which the respective element features of the multiple consecutive image frames belong, the first point cloud data, and the target registration score. Then, based on the first comprehensive score, the image frame with the highest quality can be determined.

[0069] The more types to which the element features in an image frame belong, the more corresponding point clouds, and the higher the target registration score, the higher the corresponding first comprehensive score and the higher the corresponding image quality.

[0070] The methods for determining the image point cloud data can include, but are not limited to: using a laser scanner, using a depth camera, etc., without specific limitations.

[0071] Step S310: In response to the first comprehensive score satisfying the first preset condition, determine the target image frame corresponding to the first comprehensive score as the target key frame.

[0072] The above - mentioned first preset condition can refer to a comprehensive score threshold, or can refer to the highest first comprehensive score. That is to say, after determining the first comprehensive score of each of the above - mentioned consecutive image frames, the image frame with the highest comprehensive score can be directly selected as the target key frame to improve the efficiency of target map construction; multiple image frames with relatively high comprehensive scores can also be selected as the target key frames to improve the accuracy of target map construction.

[0073] In an alternative solution of this embodiment, the system can also detect the stationary time of the vehicle. If the stationary time is too long, multiple image frames with relatively high comprehensive scores can be selected as the target key frames to repeatedly construct the initial map model corresponding to the vehicle stationary point, thereby improving the accuracy of the constructed target map.

[0074] Optionally, based on multiple image frames and the target predicted pose, determining the target registration score between each image frame and the map frame includes: obtaining the second point cloud data of each image frame and the third point cloud data of the map frame; and using a first preset algorithm to determine the target registration score between each image frame and the map frame based on the second point cloud data, the third point cloud data, and the target predicted pose.

[0075] The target registration score of an image frame can be determined by using a relevant algorithm based on the second point cloud data of the image frame and the third point cloud data of the corresponding map frame. To improve the accuracy of the determined target registration score, the third point cloud data corresponding to the map frame can also be optimized and updated by using the target predicted pose corresponding to the image frame.

[0076] In an alternative solution of this embodiment, the above-mentioned relevant algorithms may include, but are not limited to: the NDT algorithm (Normal Distributions Transform) and the ICP (Iterative Closest Point) algorithm. Preferably, the NDT algorithm is selected to calculate the target registration score of the above-mentioned image frame.

[0077] Optionally, based on the second point cloud data, the third point cloud data, and the target predicted pose, the target registration score of each image frame and the map frame is determined by using a first preset algorithm, including: updating the third point cloud data based on the target predicted pose to obtain fourth point cloud data; using a preset conversion algorithm to divide the fourth point cloud data into multiple first divided point cloud data; using a preset conversion matrix to convert the second point cloud data into second divided point cloud data corresponding to the first divided point cloud data; using the first preset algorithm to perform point cloud registration on the first divided point cloud data and the second divided point cloud data to determine the target registration score.

[0078] Taking the calculation of the above-mentioned target registration score by using the NDT algorithm as an example, specifically, first, to improve the quality of the target registration score of the calculated image frame, the map frame corresponding to the image frame can be optimized and updated by using the target predicted pose corresponding to the image frame to obtain the fourth point cloud data corresponding to the updated map frame.

[0079] Next, after obtaining the fourth point cloud data, a preset conversion algorithm can be used to divide the point cloud space corresponding to the fourth point cloud data into multiple sub-point cloud data with a specified area size according to a preset division condition, that is, the above-mentioned first divided point cloud data.

[0080] Then, according to the division condition of the fourth point cloud data, the above-mentioned image frame is also divided into multiple sub-point cloud data with a specified area size, that is, the above-mentioned second divided point cloud data.

[0081] Finally, point cloud matching is performed on the above-mentioned multiple first divided point cloud data and multiple second divided point cloud data, and the matching probability density of each specified area is calculated, so as to determine the target registration score of the image frame. The specific calculation method can refer to relevant literature and will not be elaborated here.

[0082] Optionally, multiple image frames collected by the target vehicle at different times are obtained, including: obtaining multiple target images corresponding to different times based on a preset time interval, where the preset time interval is used to represent the time difference between different times, and the multiple target images are respectively used to represent images obtained by multiple sensors of the target vehicle; splicing the multiple target images based on a second preset algorithm to obtain the image frames collected at different times.

[0083] To more accurately obtain the environmental information around the target vehicle, multiple camera devices, i.e., the above-mentioned multiple sensors, can be installed on the target vehicle. Each camera device collects multiple images containing the environmental information around the target vehicle at regular intervals, and then uses the above-mentioned second preset algorithm to splice the multiple images to form an image frame.

[0084] Specifically, four surround-view fisheye cameras can be used to collect the above-mentioned environmental information, and then the four obtained images are corrected for distortion using distortion parameters. The distortion parameters may include, but are not limited to: camera internal calibration parameters, external calibration parameters, time-axis parameters, etc., to ensure that the four surround-view fisheye cameras can collect different images at the same moment in the same world coordinate system.

[0085] After the above-mentioned images are collected, these images are transformed into multiple top-down images using homography inverse perspective transformation, and the transformed top-down images are spliced into the above-mentioned image frames.

[0086] Optionally, determining the target predicted poses of the target vehicle at different times includes: in response to the presence of a target object in the image frame, determining a first calibration result of the target object, where the first calibration result is used to represent the coordinate position of the target object in a preset map model, and the target object is used to represent an object that affects the quality of the spliced image; obtaining a second calibration result of the target vehicle, where the second calibration result is used to represent the coordinate position of the target vehicle in a preset map model; in response to the second calibration result being different from the first calibration result, determining that the target vehicle is normal at the current moment, and determining the target predicted pose at the current moment based on a pose transformation algorithm and the second actual pose of the previous frame, where the previous frame is used to represent an image frame adjacent to the current frame and obtained before the current frame; in response to the second calibration result being the same as the first calibration result, determining that the target vehicle is abnormal at the current moment and not determining the target predicted pose at the current moment.

[0087] When determining the image frame, the system can judge whether there are target objects around the target vehicle that affect the image quality based on the images collected by the camera device, such as obstacles that affect the horizontal driving of the target vehicle, such as speed bumps and stones.

[0088] If the obstacle is detected, the coordinate position of the obstacle in the preset initial map model, i.e., the first calibration result, can be determined. At the same time, the coordinate position of the target vehicle in the preset initial map model, i.e., the second calibration result, can be obtained.

[0089] If the second calibration result is the same as the first calibration result, it means that the target vehicle is currently driving to the location of the obstacle. If the target vehicle drives over the obstacle, it will cause the target vehicle to shake, which may cause the camera to capture the image with a ghost. If the image is used to construct the target map, it will greatly affect the accuracy of the constructed target map. Therefore, in order to improve the accuracy of the constructed target map, it is not determined that the image frame captured at this time is the target key frame.

[0090] If the second calibration result is the same as the first calibration result, the target predicted pose corresponding to the previous frame of the current image frame can be obtained, and then the inertial measurement value can be combined with the IMU to determine the target predicted pose corresponding to the current image frame. If the target vehicle is currently in motion, the image frame can be directly determined as the target key frame.

[0091] Figure 4 FIG. 1 is a flow chart of a semantic feature preprocessing process according to an embodiment of the present invention. Figure 4 As shown in the figure, during the target map construction process, the segmented elements generated by semantically segmenting the image frame can generally be divided into two categories: one is used to locate the target vehicle, such as the lane lines and road markings mentioned above, and the other is used to determine the image frame quality, such as obstacles such as speed bumps and rocks mentioned above. The system can detect whether the target vehicle has made contact with these obstacles. If so, the image frame is considered abnormal; if not, the image frame is considered normal, and the segmented elements used to construct the target map are determined to improve the accuracy of the constructed target map.

[0092] Optionally, after determining the target keyframe in multiple image frames based on the target registration score, the method also includes: determining the first actual pose of the target keyframe using a preset coupling algorithm based on the inertial measurement value, the first point cloud data and the target predicted pose; constructing a semantic matching factor based on the first actual pose and the target registration score, wherein the semantic matching factor is used to characterize the factor for updating the target map; judging whether the target keyframe and other keyframes form a detection loop, wherein the detection loop is used to characterize the existence of a dependency relationship between the target keyframe and the other keyframes; in response to the target keyframe and other keyframes forming a detection loop, constructing a loop detection factor based on the target keyframe and the map frame; and updating the target map based on the semantic matching factor and the loop detection factor.

[0093] After determining multiple target key frames, the preset initial map model can be preliminarily shaped based on the determined target key frames. To further improve the accuracy of the constructed target map, during the construction of the target map, the above semantic matching factor and loop detection factor can be used to optimize and update the map after preliminary shaping.

[0094] Regarding the above semantic matching factor, first, according to the above inertial measurement values, that is, parameters such as the angular velocity and acceleration of the target vehicle when collecting the above target key frame, and the first point cloud data corresponding to the target key frame, the target predicted pose corresponding to the target key frame can be coupled and corrected using a preset coupling algorithm to obtain the first actual pose corresponding to the target key frame. Then, using the first actual pose and the target registration score of the target key frame, the semantic matching factor corresponding to the target key frame can be constructed.

[0095] Regarding the above loop detection factor, first, it can be determined whether the currently determined target key frame forms a detection loop with other key frames, that is, whether the current target key frame is similar to other key frames. If there are other key frames that form a detection loop with the current target key frame, it means that the current key frame can construct the target map based on other key frames, thereby reducing the accumulated error when constructing different regions of the target map and improving the accuracy of the constructed target map; if there are no other key frames that form a detection loop with the current target key frame, continue to construct the map area corresponding to the current target key frame according to the previous key frame.

[0096] After determining that the target key frame forms a detection loop with other key frames, the loop detection factor can be further constructed based on the current target key frame and the map frame corresponding to the target key frame.

[0097] By using the semantic matching factor and loop detection factor to optimize and update the target map, it can ensure that the constructed target map is complete and smooth, closer to the real target map situation, and can effectively improve the accuracy of the constructed target map.

[0098] Optionally, determining whether the target key frame forms a detection loop with other key frames includes: obtaining the change value between the third actual pose of the target key frame and the fourth actual pose of other key frames; determining whether the change value is less than a second preset threshold; in response to the change value being less than the second preset threshold, determining whether the time interval between the target key frame and other key frames is greater than a third preset threshold; in response to the time interval being greater than the third preset threshold, determining whether the difference between the second comprehensive score of the target key frame and the third comprehensive score of other key frames is less than a fourth preset threshold; in response to the difference being less than the fourth preset threshold, determining that the target key frame forms a detection loop with other key frames.

[0099] The method for determining whether the current target key frame forms a loop detection with other key frames is as follows:

[0100] First, the actual pose change value between the target key frame and other key frames can be obtained, and it is judged whether this change value is less than the preset pose threshold, that is, the above-mentioned second preset threshold. If it is less, it means that the target key frame is similar to other key frames and may form a detection loop; if it is not less, it means that the target key frame is different from other key frames and does not form a detection loop.

[0101] Next, if the above actual pose change value is small, it is further judged whether the time interval between the current target key frame and the above other key frames is greater than the preset time threshold, that is, the above-mentioned third time threshold. If it is greater, it means that the acquisition times of the target key frame and other key frames are quite different and may form a detection loop; if it is not greater, it means that the acquisition times of the target key frame and other key frames are too close, which may be because the current driving speed of the target vehicle is too slow or the angular rate is too small, resulting in the target key frame and other key frames being images of the same environmental scene.

[0102] Then, if the above time interval is large, it is further judged whether the comprehensive score of the current target key frame and the above other key frames is greater than the preset score threshold, that is, the above-mentioned fourth preset threshold. If it is greater, it is determined that the target key frame and other key frames form a detection loop; if it is not greater, it is determined that the target key frame and other key frames do not form a detection loop.

[0103] To clearly show the above method of map construction, Figure 5 is a schematic flowchart of a local map construction shown according to an embodiment of the present invention. As Figure 5 shown, after obtaining the segmentation element information of the image frame, that is, the number of types of segmentation elements and the corresponding point cloud data, the IMU data of the target vehicle, that is, the inertial measurement value, and the pose of the previous image frame can be used simultaneously to determine the target predicted pose of the current image frame; then, according to the number of types of the above segmentation elements, the point cloud data of the image frame, and the target predicted pose, the NDT algorithm is used to determine the current actual pose and the target registration score of the image frame; then, according to the above inertial measurement value and the driving speed of the target vehicle, it is judged whether the target vehicle is in a moving state. If it is, the image frame can be directly determined as the target key frame; if it is not, the image frame is stored in the post-processing set, and when the target vehicle is in a moving state again, a comprehensive score is performed according to the target registration score, the number of types of segmentation elements, and the point cloud data of each image frame in the set, and the one with the highest comprehensive score is determined as the target key frame in the set; finally, after determining the target key frame, the target map can be constructed based on these target key frames.

[0104] Embodiment 2

[0105] According to another aspect of the embodiments of the present invention, corresponding to the embodiments of the fault diagnosis of the above charging device, this specification also provides a fault diagnosis device for a charging device. Please refer to Figure 6 , Figure 6 is a structural block diagram of a map construction device shown according to an embodiment of the present invention. The device includes: a first acquisition module 602, configured to acquire multiple image frames collected by a target vehicle at different times, where the image frames are used to characterize the environment around the target vehicle; a first determination module 604, configured to determine the target predicted poses of the target vehicle at different times; a second determination module 606, configured to determine the target registration scores of each image frame and a map frame based on the multiple image frames and the target predicted poses, where the map frame is an image frame corresponding to a preset map model acquired by the target vehicle at different times; a third determination module 608, configured to determine target key frames among the multiple image frames based on the target registration scores; and a map update module 610, configured to process the target key frames using the preset map model to obtain a target map.

[0106] Optionally, the third determination module 608 includes: a segmentation element determination unit, configured to perform semantic segmentation on each image frame to obtain segmentation elements of each image, where the segmentation elements are used to characterize the recognition objects included in each image frame; a first acquisition unit, configured to acquire the target speed and inertial measurement values of the target vehicle, where the target speed is used to characterize the driving speed of the target vehicle at different times, and the inertial measurement values are used to characterize the angular rate and acceleration of the target vehicle at different times; a first storage unit, configured to determine that the target vehicle is stationary in response to the target speed and inertial measurement values being less than a first preset threshold, and store the target image frame corresponding to the stationary target vehicle into a post-processing frame set, where the post-processing frame set is a set composed of image frames continuously acquired when the target vehicle is stationary; a comprehensive score determination unit, configured to determine a first comprehensive score of each target image frame in the post-processing frame set based on the number of different types to which the segmentation elements corresponding to any target image frame in the post-processing frame set belong, first point cloud data, and the target registration scores; and a target key frame determination unit, configured to determine the target image frame corresponding to the first comprehensive score as the target key frame in response to the first comprehensive score satisfying a first preset condition.

[0107] Optionally, the second determination unit includes: a second acquisition unit, configured to acquire second point cloud data of each image frame and third point cloud data of the map frame; and a registration score determination unit, configured to determine the target registration scores of each image frame and the map frame based on the second point cloud data, the third point cloud data, and the target predicted poses using a first preset algorithm.

[0108] Optionally, the registration score determination unit includes: a first determination subunit, configured to update the third point cloud data based on the target predicted pose to obtain fourth point cloud data; a first division subunit, configured to divide the fourth point cloud data into multiple first divided point cloud data by using a preset conversion algorithm; a first conversion subunit, configured to convert the second point cloud data into second divided point cloud data corresponding to the first divided point cloud data by using a preset conversion matrix; and a first registration subunit, configured to perform point cloud registration on the first divided point cloud data and the second divided point cloud data by using a first preset algorithm to determine a target registration score.

[0109] Optionally, the first acquisition module 602 includes: a target image acquisition unit, configured to acquire multiple target images corresponding to different moments based on a preset time interval, where the preset time interval is used to represent the time difference between different moments, and the multiple target images are respectively used to represent images acquired by multiple sensors based on a target vehicle; and an image stitching unit, configured to stitch the multiple target images based on a second preset algorithm to obtain an image frame acquired at different moments.

[0110] Optionally, the first determination module 604 includes: a first calibration determination unit, configured to determine a first calibration result of a target object in response to the presence of the target object in the image frame, where the first calibration result is used to represent the coordinate position of the target object in a preset map model, and the target object is used to represent an object affecting the quality of the stitched image; a second calibration acquisition unit, configured to acquire a second calibration result of the target vehicle, where the second calibration result is used to represent the coordinate position of the target vehicle in the preset map model; and a target predicted pose determination unit, configured to determine that the target vehicle is normal at the current moment in response to the second calibration result being different from the first calibration result, and determine the target predicted pose at the current moment based on a pose transformation algorithm and the second actual pose of the previous frame, where the previous frame is used to represent an image frame adjacent to the current frame and acquired before the current frame; and determine that the target vehicle is abnormal at the current moment and not determine the target predicted pose at the current moment in response to the second calibration result being the same as the first calibration result.

[0111] Optionally, the device further includes: an actual pose determination module, configured to determine a first actual pose of a target key frame based on inertial measurement values, first point cloud data, and a target predicted pose by using a preset coupling algorithm; a matching factor construction module, configured to construct a semantic matching factor based on the first actual pose and a target registration score, where the semantic matching factor is used to represent a factor for updating a target map; a loop detection and judgment module, configured to judge whether the target key frame and other key frames form a detection loop, where the detection loop is used to represent a dependency relationship between the target key frame and other key frames; a detection factor construction module, configured to, in response to the target key frame and other key frames forming a detection loop, construct a loop detection factor based on the target key frame and a map frame; and a target map update module 610, configured to update the target map based on the semantic matching factor and the loop detection factor.

[0112] Optionally, the loop detection and judgment module includes: a third acquisition unit, configured to acquire a change value between a third actual pose of the target key frame and a fourth actual pose of other key frames; a first judgment unit, configured to judge whether the change value is less than a second preset threshold; a second judgment unit, configured to, in response to the change value being less than the second preset threshold, judge whether a time interval between the target key frame and other key frames is greater than a third preset threshold; a third judgment unit, configured to, in response to the time interval being greater than the third preset threshold, judge whether a difference between a second comprehensive score of the target key frame and a third comprehensive score of other key frames is less than a fourth preset threshold; and a loop determination unit, configured to, in response to the difference being less than the fourth preset threshold, determine that the target key frame and other key frames form a detection loop.

[0113] Embodiment 3

[0114] On the other hand, according to an embodiment of the present invention, there is also provided a computer-readable storage medium, where the computer-readable storage medium includes a stored program, and when the program runs, it controls a device where the computer-readable storage medium is located to execute the map construction method in the above method embodiment.

[0115] Embodiment 4

[0116] On the other hand, according to an embodiment of the present invention, there is also provided a processor, where the processor is used to run a program, and when the program runs, it executes the map construction method in the above method embodiment.

[0117] Embodiment 5

[0118] On the other hand, according to an embodiment of the present invention, there is also provided a target vehicle, including: one or more processors; a storage device, configured to store one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors execute the map construction method in any one of the above.

[0119] The serial numbers of the above embodiments of the present invention are only for description and do not represent the superiority or inferiority of the embodiments.

[0120] In the above embodiments of the present invention, the descriptions of the respective embodiments have their own emphases. For the parts not detailed in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0121] In the several embodiments provided by the present application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only illustrative. For example, the division of units can be a logical function division. In actual implementation, there can be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection to each other can be through some interfaces. The indirect coupling or communication connection of units or modules can be in an electrical or other form.

[0122] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0123] In addition, the functional units in the various embodiments of the present invention can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0124] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the methods in the various embodiments of the present invention. And the aforementioned storage medium includes: USB flash drives, read-only memories (ROMs), random access memories (RAMs), mobile hard disks, magnetic disks or optical disks and other various media that can store program codes.

[0125] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A method for map construction, characterized in that, Including: Obtaining multiple image frames collected by a target vehicle at different times, where the image frames are used to represent the environment around the target vehicle; Determining the target predicted poses of the target vehicle at the different times; Based on the multiple image frames and the target predicted poses, determining the target registration scores of each image frame with a map frame, where the map frame is used to represent the image frame corresponding to a preset map model obtained by the target vehicle at the different times, and the target registration score of each image frame with the map frame is determined by using a first preset algorithm based on the second point cloud data of each image frame, the third point cloud data of the map frame, and the target predicted poses, and the first preset algorithm includes at least one of the following: normal transformation distribution algorithm and iterative closest point algorithm; Based on the target registration scores, determining the target key frames in the multiple image frames, where the target key frame is the target image frame corresponding to a first comprehensive score when the first comprehensive score meets a first preset condition, the first comprehensive score is used to represent the comprehensive scores of each target image frame in a post-processing frame set, the first comprehensive score is determined by the number of different types of segmentation elements corresponding to any target image frame in the post-processing frame set, the first point cloud data, and the target registration scores, the post-processing frame set is a set composed of image frames continuously obtained when the target vehicle is stationary, the segmentation element is an element obtained by performing semantic segmentation on each image frame, and the segmentation element is used to represent the recognition object included in each image frame; Processing the target key frames by using the preset map model to obtain a target map.

2. The method according to claim 1, wherein The method further includes: Obtaining the target speed and inertial measurement values of the target vehicle, where the target speed is used to represent the driving speed of the target vehicle at the different times, and the inertial measurement values are used to represent the angular rate and acceleration of the target vehicle at the different times; In response to the target speed and the inertial measurement values being less than a first preset threshold, determining that the target vehicle is stationary, and storing the target image frames corresponding to the stationary target vehicle into the post-processing frame set.

3. The method according to claim 1, wherein Based on the second point cloud data, the third point cloud data, and the target predicted poses, using the first preset algorithm to determine the target registration scores of each image frame with the map frame, including: Updating the third point cloud data based on the target predicted poses to obtain a fourth point cloud data; Using a preset conversion algorithm to divide the fourth point cloud data into multiple first divided point cloud data; Using a preset conversion matrix to convert the second point cloud data into second divided point cloud data corresponding to the first divided point cloud data; Using the first preset algorithm to perform point cloud registration on the first divided point cloud data and the second divided point cloud data to determine the target registration scores.

4. The method according to claim 1, wherein Obtaining multiple image frames collected by a target vehicle at different times, including: Obtain multiple target images corresponding to the different moments based on a preset time interval, where the preset time interval is used to represent the time difference between the different moments, and the multiple target images are respectively used to represent the images acquired by multiple sensors based on which the target vehicle is; Stitch the multiple target images based on a second preset algorithm to obtain the image frame acquired at the different moments.

5. The method according to claim 4, wherein Determine the target predicted poses of the target vehicle at the different moments, including: In response to the existence of a target object in the image frame, determine a first calibration result of the target object, where the first calibration result is used to represent the coordinate position of the target object in the preset map model, and the target object is used to represent an object that affects the normal driving of the target vehicle; Obtain a second calibration result of the target vehicle, where the second calibration result is used to represent the coordinate position of the target vehicle in the preset map model; In response to the second calibration result being different from the first calibration result, determine that the target vehicle is normal at the current moment, and based on a pose transformation algorithm and the second actual pose of the previous frame, determine the target predicted pose at the current moment, where the previous frame is used to represent an image frame adjacent to the current frame and acquired before the current frame; In response to the second calibration result being the same as the first calibration result, determine that the target vehicle is abnormal at the current moment and do not determine the target predicted pose at the current moment.

6. The method according to claim 2, wherein After determining the target key frames in the multiple image frames based on the target registration score, the method further includes: Based on the inertial measurement value, the first point cloud data, and the target predicted pose, use a preset coupling algorithm to determine the first actual pose of the target key frame; Based on the first actual pose and the target registration score, construct a semantic matching factor, where the semantic matching factor is used to represent a factor for updating the target map; Determine whether the target key frame and other key frames form a detection loop, where the detection loop is used to represent a dependency relationship between the target key frame and the other key frames; In response to the target key frame and the other key frames forming a detection loop, construct a loop detection factor based on the target key frame and the map frame; Update the target map based on the semantic matching factor and the loop detection factor.

7. The method according to claim 6, characterized in that, Determining whether the target key frame and other key frames form a detection loop includes: Obtain the change value between the third actual pose of the target key frame and the fourth actual pose of the other key frames; Determine whether the change value is less than a second preset threshold; In response to the change value being less than the second preset threshold, determine whether the time interval between the target key frame and the other key frames is greater than a third preset threshold; In response to the time interval being greater than the third preset threshold, determine whether the difference between the second comprehensive score of the target key frame and the third comprehensive score of the other key frames is less than a fourth preset threshold; In response to the difference being less than the fourth preset threshold, determine that the target key frame and the other key frames form a detection loop.

8. A map construction device, characterized in that, Include: A first acquisition module, configured to acquire multiple image frames collected by a target vehicle at different times, where the image frames are used to represent the environment around the target vehicle; A first determination module, configured to determine the target predicted poses of the target vehicle at the different times; A second determination module, configured to determine, based on the multiple image frames and the target predicted poses, a target registration score between each image frame and a map frame, where the map frame is used to represent an image frame corresponding to a preset map model acquired by the target vehicle at the different times, and the target registration score between each image frame and the map frame is determined by using a first preset algorithm based on the second point cloud data of each image frame, the third point cloud data of the map frame, and the target predicted poses, and the first preset algorithm includes at least one of the following: normal transformation distribution algorithm and iterative closest point algorithm; A third determination module, configured to determine, based on the target registration scores, target key frames among the multiple image frames, where the target key frames are target image frames corresponding to a first comprehensive score when the first comprehensive score meets a first preset condition, the first comprehensive score is used to represent the comprehensive score of each target image frame in a post-processing frame set, the first comprehensive score is determined by the number of different types of segmentation elements corresponding to any target image frame in the post-processing frame set, the first point cloud data, and the target registration scores, the post-processing frame set is a set composed of image frames continuously acquired when the target vehicle is stationary, the segmentation element is an element obtained by performing semantic segmentation on each image frame, and the segmentation element is used to represent the recognition object included in each image frame; A map update module, configured to process the target key frames by using the preset map model to obtain a target map.

9. A vehicle, comprising: At least one processor; And A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is enabled to execute the method according to any one of claims 1-7.

Images