Method, system, mapping method and medium for extracting coordinates of underground parking elements

Abstract

The present application discloses a method, system, mapping method and medium for extracting coordinates of underground parking elements, which belongs to the field of geographic information and image technology. The method includes: collecting environmental data of the underground parking lot, and determining the longitude and latitude coordinates of the reference points in the environmental data; converting the environmental data, and segmenting the obtained initial image to obtain a tile map; marking the parking lot elements in the tile map, and processing the tile map using a deep learning algorithm model to obtain the polygonal outline of the parking lot elements; fusing multiple tile maps, and in the fused image, determining the longitude and latitude coordinates corresponding to the polygonal outline according to the position and longitude and latitude coordinates of the reference points. The present application determines the longitude and latitude coordinates of the environmental data of the underground parking lot, so that the collected environmental data are interconnected, thereby reducing the production cost and speeding up the parking lot map production process when mapping the parking spaces and pillars of the underground parking lot.

Description

Technical Field

[0001] The present application relates to the fields of geographic information and image technology, and in particular to a method system, a mapping method, and a medium for extracting coordinates of elements in an underground parking lot. Background Art

[0002] When it comes to identifying parking spaces in parking lots, existing methods primarily fall into two categories: non-video image detection and video image recognition. Non-video image detection techniques include induction coil detection, acoustic wave detection, infrared detection, and RFID-based radio frequency identification technology. Video image recognition techniques primarily utilize traditional image detection algorithms and deep learning algorithms.

[0003] Among them, when using non-video image detection technologies for parking space detection, installation of these sensors requires modifications to the parking lot surface, making installation difficult and costly, significantly increasing the complexity of equipment maintenance and repair. Using traditional image detection algorithms for parking space recognition: When using video image recognition methods for parking space recognition, traditional image algorithms have poor generalization and robustness, are highly susceptible to interference from environmental changes, and have poor parking space recognition capabilities. Furthermore, current patents or solutions using deep learning algorithms for parking space recognition use only video or image data as input, sourced from cameras. These images are then used for parking space recognition and classification, with the output being either "occupied" or "unoccupied." Because the deep learning algorithm's raw data lacks GPS positioning information and the video image data lacks spatial correlation, it is impossible to map parking space information into a complete parking lot map, hindering the creation and updating of comprehensive parking lot maps. Summary of the Invention

[0004] In response to the technical problems in the existing technology of high cost, poor accuracy and incomplete recognition results of parking spaces and pillars in parking lots, this application proposes a method, system, mapping method and medium for extracting coordinates of underground parking lot elements.

[0005] In the first aspect, the present application provides a method for extracting the coordinates of underground parking lot elements, including: collecting environmental data of the underground parking lot and determining the latitude and longitude coordinates of the reference point in the environmental data; converting the environmental data and segmenting the obtained initial image to obtain multiple tile images; marking the parking lot elements in the tile image, and processing the tile image using a pre-trained deep learning algorithm model to obtain the polygonal outlines corresponding to the parking lot elements in each tile image; and fusing the processed multiple tile images to obtain a fused image, and in the fused image, determining the latitude and longitude coordinates corresponding to the polygonal outline of the parking lot element based on the relative position of the polygonal outline of the parking lot element and the reference point and the latitude and longitude coordinates of the reference point.

[0006] Optionally, a pre-trained deep learning algorithm model is used to process the tile map to obtain the polygonal outlines corresponding to the parking lot elements in each tile map, including: performing a downsampling convolution operation on the tile map to extract the feature map in the tile map; processing the feature map through a clustering dimensionality reduction algorithm to obtain the detection frame corresponding to the feature map; performing a deconvolution operation on the feature map, and performing regression processing on the detection frame in the feature map after the deconvolution operation to determine the polygonal outline of the parking lot element.

[0007] Optionally, the tile map is processed using a pre-trained deep learning algorithm model to obtain polygonal outlines corresponding to parking lot elements in each tile map. The method also includes: before performing a downsampling convolution operation on the tile map, amplifying the tile map to obtain a multi-dimensional tile map, and the amplification includes angle rotation; before performing a deconvolution operation on the feature map, multiple corresponding multi-dimensional tile maps are fused to enhance the features in the tile map.

[0008] Optionally, in the fused image, the longitude and latitude coordinates corresponding to the outline of the parking lot element are determined based on the relative positions of the polygonal outline of the parking lot element and the reference point and the longitude and latitude coordinates of the reference point, including: in the fused point image, obtaining the pixel coordinates of the outline of the parking lot element and the pixel coordinates of the reference point, and then determining the relative position of the outline of the parking lot element and the reference point; converting the longitude and latitude coordinates of the reference point according to the relative position to obtain the longitude and latitude coordinates corresponding to the polygonal outline of the parking lot element.

[0009] Optionally, in the fused image, the longitude and latitude coordinates corresponding to the outline of the parking lot element are determined based on the relative position of the outline of the parking lot element and the reference point and the longitude and latitude coordinates of the reference point, and the method also includes: performing a vectorization operation on the polygonal outline of the parking lot element in the fused image, correcting the polygonal outline, and obtaining a rectangular outline of the parking lot element; obtaining the pixel coordinates of the rectangular outline and the pixel coordinates of the reference point, and then determining the relative position of the rectangular outline and the reference point.

[0010] Optionally, segmenting the initial image to obtain a plurality of tile images includes segmenting the initial image to obtain a plurality of tile images according to a preset ratio of overlap between adjacent tile images.

[0011] Optionally, the environmental data is converted, including: under the condition that the environmental data is point cloud data, selecting the point cloud data according to the elevation information of the point cloud data to determine the parking lot point cloud data and the pillar point cloud data; converting the parking lot point cloud data and the pillar point cloud data respectively to obtain an initial image including the parking lot point cloud map and the pillar point cloud map.

[0012] Optionally, segmenting the initial image to obtain a plurality of tile images includes: determining a segmentation range according to the size of parking spaces in the underground parking lot; and segmenting the initial image according to the segmentation range to obtain a plurality of tile images.

[0013] In the second aspect, the present application provides a system for extracting the coordinates of underground parking lot elements, including: a data acquisition module, which collects environmental data of the underground parking lot and determines the latitude and longitude coordinates of the reference point in the environmental data; an image conversion and segmentation module, which converts the environmental data and segments the obtained initial image to obtain multiple tile images; a contour determination module, which marks the parking lot elements in the tile image and uses a pre-trained deep learning algorithm model to process the tile image to obtain the polygonal contours corresponding to the parking lot elements in each tile image; and a coordinate determination module, which fuses the processed multiple tile images to obtain a fused image, and in the fused image, determines the latitude and longitude coordinates corresponding to the polygonal contour of the parking lot element based on the relative position of the polygonal contour of the parking lot element and the reference point and the latitude and longitude coordinates of the reference point.

[0014] On the third aspect, the present application provides a method for producing an underground parking lot map, including: collecting environmental data of the underground parking lot and determining the latitude and longitude coordinates of the reference point in the environmental data; converting the environmental data and segmenting the obtained initial image to obtain multiple tile images; marking the parking lot elements in the tile image, and using a pre-trained deep learning algorithm model to process the tile image to obtain the polygonal outlines corresponding to the parking lot elements in each tile image; fusing the processed multiple tile images to obtain a fused image, and in the fused image, determining the latitude and longitude coordinates corresponding to the polygonal outline of the parking lot element based on the relative position of the polygonal outline of the parking lot element and the reference point and the latitude and longitude coordinates of the reference point; establishing an underground parking lot map based on the polygonal outline and the latitude and longitude coordinates corresponding to the polygonal outline.

[0015] In a fourth aspect, the present application provides a computer-readable storage medium storing computer instructions, wherein the computer instructions are operated to execute the method for extracting coordinates of underground parking elements in Scheme 1 or the method for producing an underground parking map in Scheme 3.

[0016] The beneficial effect of this application is that this application determines the latitude and longitude coordinates of the underground parking lot environmental data, so that the collected environmental data are interconnected, thereby reducing production costs and speeding up the parking lot map production process when mapping the parking spaces and pillars in the underground parking lot. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0018] Figure 1 This is a flowchart of a specific implementation method of the method for extracting coordinates of underground parking elements in this application;

[0019] Figure 2 This is a schematic diagram of the operation flow of an example of the deep learning instance segmentation algorithm model of this application;

[0020] Figure 3 It is a flowchart of the vectorization operation in the method for extracting coordinates of underground parking elements in this application;

[0021] Figure 4 This is a flowchart of an example of point cloud tile cutting and splicing in the method of extracting parking spaces and column coordinates in the underground parking lot in this application;

[0022] Figure 5 It is a schematic diagram of a specific embodiment of the system for extracting coordinates of underground parking elements of the present application;

[0023] Figure 6 It is a schematic diagram of an implementation method of the underground parking lot map making method of the present application.

[0024] The above drawings illustrate specific embodiments of the present application, which will be described in more detail below. These drawings and the textual description are not intended to limit the scope of the present application in any way, but rather to illustrate the concepts of the present application to those skilled in the art by reference to specific embodiments. DETAILED DESCRIPTION

[0025] To make the purpose, technical solutions, and advantages of the embodiments of this application more clear, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0026] The terms "first," "second," "third," "fourth," and the like (if any) in the specification and claims of the present application and in the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or sequential sequence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of the present application described herein can, for example, be implemented in orders other than those illustrated or described herein. In addition, the terms "including" and "having," and any variations thereof, are intended to cover non-exclusive inclusions, e.g., a product or apparatus comprising a series of steps or units is not necessarily limited to those units explicitly listed, but may include other units not explicitly listed or inherent to those products or apparatuses.

[0027] Existing parking space identification methods primarily fall into two categories: non-video image detection and video image recognition. Non-video image detection techniques primarily include induction coil detection, acoustic wave detection, infrared detection, and RFID-based radio frequency identification technology. Video image recognition techniques primarily include traditional image detection algorithms and deep learning algorithms. Non-video image detection techniques require modifications to the parking lot floor, making installation difficult and costly, significantly increasing equipment maintenance and repair. Traditional image detection algorithms suffer from poor generalization and robustness, are susceptible to interference from environmental changes, and exhibit limited parking space recognition capabilities. Furthermore, current patents and solutions using deep learning algorithms for parking space identification rely solely on video or image data from cameras as input. Parking space classification is then performed on the images, with the output being either "occupied" or "empty." Since the original data of the deep learning algorithm does not contain GPS positioning information and there is no spatial information correlation between the video image data, it is impossible to draw the parking space information into a complete parking lot map, and it is impossible to produce and update a complete parking lot map.

[0028] Therefore, in response to the technical problems existing in the prior art, the present application provides a method, system, mapping method and medium for extracting the coordinates of underground parking lot elements. The method for extracting the coordinates of underground parking lot elements includes: collecting environmental data of the underground parking lot and determining the longitude and latitude coordinates of the reference point in the environmental data; converting the environmental data and segmenting the obtained initial image to obtain multiple tile images; marking the parking lot elements in the tile image and processing the tile image using a pre-trained deep learning algorithm model to obtain the polygonal outlines corresponding to the parking lot elements in each tile image; and fusing the multiple tile images to obtain a fused image, and in the fused image, determining the longitude and latitude coordinates corresponding to the polygonal outline of the parking lot element based on the relative position of the polygonal outline of the parking lot element and the reference point and the longitude and latitude coordinates of the reference point.

[0029] This application collects environmental data from underground parking lots, including image data with GPS information or point cloud data obtained by scanning the entire parking lot using a lidar. The converted initial image is recognized and processed using a deep learning algorithm, then vectorized and integrated with longitude and latitude information to convert the outline information of each parking space and column in the parking lot into longitude and latitude coordinates. The ability to quickly convert parking space and column information in the parking lot into map data greatly reduces the workload of front-line production workers, reduces production costs, and speeds up the production and updating of parking lot maps.

[0030] The following describes in detail the technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems using specific embodiments. The specific embodiments described below can be combined with each other to form new embodiments. The same or similar ideas or processes described in one embodiment may not be repeated in other embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0031] Figure 1 An embodiment of the method for extracting coordinates of underground parking elements of the present application is shown.

[0032] exist Figure 1 In the illustrated embodiment, the method for extracting coordinates of underground parking elements of the present application includes a process S101 of collecting environmental data of the underground parking lot and determining the longitude and latitude coordinates of reference points in the environmental data.

[0033] In this embodiment, because the existing technology lacks GPS information in the underground parking lot data collected during underground parking lot map creation, there is no spatial correlation between the collected video image data, making it impossible to create an underground parking lot map. In the process of collecting underground parking lot environmental data, this application uses image data with GPS information, or scans the underground parking lot using a laser radar (LiDAR) to obtain point cloud data corresponding to the parking lot. The point cloud data corresponding to the parking lot includes point cloud data for each parking space in the underground parking lot and point cloud data for each pillar in the underground parking lot. By adding GPS information to the image or acquiring point cloud data in the underground parking lot using a laser radar, in addition to processing the point cloud data to obtain the corresponding underground parking lot image information, the point cloud data acquired by the laser radar also includes the location information of each point cloud. After obtaining image information of the parking spaces and pillars in the underground parking lot, the corresponding latitude and longitude coordinate information can also be obtained. This provides image information and location information for the parking spaces and pillars in the subsequent underground parking lot map creation process, expediting the creation of the parking lot map.

[0034] Specifically, when acquiring image data with GPS information, an industrial synthetic camera can be used to directly or indirectly obtain the longitude and latitude information of the pixel points in the image through the precise GPS positioning function while taking pictures, and obtain the longitude and latitude coordinates of each pixel point in the image. When acquiring point cloud image data, the underground parking lot can be scanned by equipment such as a laser radar to obtain point cloud data of the underground parking lot. In addition, when selecting reference points in environmental data, the reference points need to have clear longitude and latitude coordinates so that the positions of other points can be determined through the reference points. For example, the middle point in the image data or point cloud data with GPS information can be used as the reference point. It should be noted that the reference point can be reasonably selected based on the actual data collected, as long as the reference point has accurate longitude and latitude coordinates. This application does not impose specific limits on the selection of specific reference points.

[0035] exist Figure 1 In the illustrated embodiment, the method for extracting coordinates of underground parking elements of the present application includes: process S102, converting environmental data and segmenting the obtained initial image to obtain multiple tile images.

[0036] In this embodiment, after obtaining environmental data of the underground parking lot, such as point cloud data of the underground parking lot through a lidar, the point cloud data of the underground parking lot needs to be pre-processed to convert the point cloud data of the underground parking lot into a point cloud image of the underground parking lot. The image is then segmented to obtain a subsequent tile image.

[0037] Optionally, the environmental data is converted, including: under the condition that the environmental data is point cloud data, selecting the point cloud data according to the elevation information of the point cloud data to determine the parking lot point cloud data and the pillar point cloud data; converting the parking lot point cloud data and the pillar point cloud data respectively to obtain an initial image including the parking lot point cloud map and the pillar point cloud map.

[0038] In this optional embodiment, the conversion process of the collected environmental data of the underground parking lot is mainly for point cloud data, converting the point cloud data into image data. The conversion process can be omitted for the collected image data with GPS information. When processing the point cloud data of the underground parking lot, the point cloud data of the underground parking lot mainly includes parking space point cloud data and column point cloud data. Therefore, first, based on the distribution and location characteristics of the parking spaces and columns, the point cloud data is divided into column point cloud data and parking space point cloud data according to the elevation information of the point cloud data. Then, the column point cloud image corresponding to the column is generated based on the column point cloud data, and the corresponding parking space point cloud image is generated based on the parking space point cloud data, thereby obtaining the initial image corresponding to the point cloud data.

[0039] Specifically, when dividing the point cloud data into column point cloud data and parking point cloud data based on the elevation information of the point cloud data of the underground parking lot, according to the characteristics of the columns and parking spaces, the height experience value of 1.5 meters can be selected as the standard for dividing the parking point cloud data and the column point cloud data. Among them, the point cloud data above 1.5 meters is divided into column point cloud data, and the point cloud data below 1.5 meters is divided into parking point cloud data. It should be noted that when dividing the actual column point cloud data and the parking point cloud data, the appropriate division standard can be set according to the positional relationship between the parking spaces and the columns in different underground parking lots in the actual scene. The standard of 1.5 meters can be selected for division, or other suitable values ​​can be selected. This application does not impose specific restrictions on the specific value of the height.

[0040] Specifically, in the process of generating corresponding column point cloud images and parking point cloud images based on the column point cloud data and the parking point cloud data, the parking point cloud image mainly includes the parking space contour lines, column contour lines, lane lines, car contours and other height standards on the ground, such as the point cloud information of all objects below 1.5 meters; the column point cloud image includes the column contours and wall contours in the underground parking lot.

[0041] Specifically, the preprocessing of the point cloud data of the underground parking lot acquired by the LiDAR includes filtering and noise reduction. This filtering eliminates some point cloud data other than those of parking spaces and pillars to ensure the accuracy of the point cloud data. This helps to avoid the influence of impurity point clouds during the subsequent segmentation of the point cloud data and the generation of point cloud images for parking spaces and pillars, thereby improving the accuracy of these images.

[0042] By dividing the underground parking lot point cloud data by elevation, corresponding parking spot point cloud images and pillar point cloud images are generated. Impure point cloud data is filtered out through point cloud data processing, including but not limited to noise reduction and filtering, to ensure the accuracy of the obtained parking spot point cloud images and pillar point cloud images.

[0043] Optionally, segmenting the initial image to obtain a plurality of tile images includes: determining a segmentation range according to the size of parking spaces in the underground parking lot; and segmenting the initial image according to the segmentation range to obtain a plurality of tile images.

[0044] In this optional embodiment, the initial image is segmented to obtain smaller tiles to facilitate feature extraction in the deep learning algorithm model. To avoid cutting into a complete parking space, which could cause errors in subsequent feature recognition, the segmentation range of the initial image is determined based on the size of the parking space to facilitate subsequent processing.

[0045] Specifically, the point cloud image of the underground parking lot obtained directly is large and not suitable for the needs of the deep learning instance segmentation algorithm model. Therefore, the initial image is cut according to the range of parking spaces in the underground parking lot to obtain multiple tile images.

[0046] Specifically, based on the needs of the deep learning instance segmentation algorithm model and the actual length and width of the parking spaces in the parking lot, the larger underground parking lot point cloud image is cut into 1536*1536 tile images. Subsequently, multiple tile images are subjected to data annotation, manual quality inspection, format conversion and other preprocessing operations to produce a data format suitable for the deep learning algorithm model.

[0047] Specifically, when segmenting the initial image of an underground parking lot, for example, the parking space and pillar point cloud images within the corresponding point cloud data are segmented to generate multiple point cloud tiles. Each parking space and pillar outline in the tiled point cloud can be marked with a polygon. The polygon coordinates are saved in the corresponding file.

[0048] Optionally, segmenting the initial image to obtain a plurality of tile images includes segmenting the initial image to obtain a plurality of tile images according to a preset ratio of overlap between adjacent tile images.

[0049] In this embodiment, when the entire parking lot image, such as a parking spot cloud image or a column point cloud image, is sliced, the parking spaces are cut in half, making it impossible for the algorithm model to detect a complete parking space outline. Therefore, when formulating the cutting strategy, a semi-covering cropping method is used to segment the entire parking lot image. This ensures that each point cloud tile maintains a preset ratio of overlap with its adjacent point cloud tiles.

[0050] Specifically, when dividing the parking lot image into tiles, each tile is set to have a 1 / 3 overlap with adjacent tiles. Adjacent tiles include the overlapping portions of the tile above, below, left, and right. This ensures that even if a parking space is cut in half at the edge of the previous tile, the next tile will still contain a complete parking space. This ensures that the tile contains complete parking spaces, ensuring data accuracy.

[0051] exist Figure 1 In the illustrated embodiment, the method of extracting coordinates of underground parking lot elements of the present application includes: process S103, marking the parking lot elements in the tile map, and processing the tile map using a pre-trained deep learning algorithm model to obtain polygonal contours corresponding to the parking lot elements in each tile map.

[0052] In this implementation, parking lot elements include parking spaces and pillars, which are the primary elements for subsequent parking lot mapping. After obtaining the tile image, the parking spaces and pillars are marked using their characteristics. For example, in a point cloud image, the point cloud density of pillars will be much higher than that of other background objects. The tile image is then processed using a deep learning algorithm model to ultimately determine the outlines of the parking spaces and pillars that are located in the tile image.

[0053] Optionally, the parking lot elements in the tile map are marked, including: labeling the parking space data and column data in the tile map, and marking the parking space outlines and column outlines corresponding to the parking space data and column data respectively through polygons to obtain the corresponding parking space polygon outlines and column polygon outlines; determining the pixel coordinates corresponding to the parking space polygon outline and the column polygon outline, and storing them.

[0054] In this optional embodiment, after marking the parking spaces and columns in the tile image, polygons are used to mark the parking space outlines and column outlines to obtain the corresponding parking space outlines and column outlines. This marking of the outlines facilitates the processing of the deep learning algorithm model and improves accuracy. Simultaneously, the pixel coordinates corresponding to the parking space outlines and column outlines are recorded in the tile image. The recording of pixel coordinates also records the pixel positions of the parking space outlines and column outlines, facilitating the subsequent determination of the actual positions of the parking spaces and columns.

[0055] Optionally, a pre-trained deep learning algorithm model is used to process the tile map to obtain the polygonal outlines corresponding to the parking lot elements in each tile map, including: performing a downsampling convolution operation on the tile map to extract the feature map in the tile map; processing the feature map through a clustering dimensionality reduction algorithm to obtain the detection frame corresponding to the feature map; performing a deconvolution operation on the feature map, and performing regression processing on the detection frame in the feature map after the deconvolution operation to determine the polygonal outline of the parking lot element.

[0056] In this optional embodiment, after the parking lot elements in the tile image are labeled, the tile image is fed into a deep learning algorithm model for processing, ultimately obtaining the outlines of the parking spaces and pillars. During the deep learning algorithm model's processing of the tile image, a downsampling convolution operation is performed on the tile image to extract a feature map from the tile image; the feature map is processed using a clustering dimensionality reduction algorithm to obtain a detection box corresponding to the feature map; a deconvolution operation is performed on the feature map, and regression processing is performed on the detection box in the deconvolution feature map to determine the polygonal outline of the parking lot elements.

[0057] Optionally, the tile map is processed using a pre-trained deep learning algorithm model to obtain polygonal outlines corresponding to parking lot elements in each tile map. The method also includes: before performing a downsampling convolution operation on the tile map, amplifying the tile map to obtain a multi-dimensional tile map, and the amplification includes angle rotation; before performing a deconvolution operation on the feature map, multiple corresponding multi-dimensional tile maps are fused to enhance the features in the tile map.

[0058] In this optional embodiment, in order to improve the processing accuracy of the deep learning algorithm model, the number of tile images will be expanded, such as angle rotation, mirror flipping or random cropping of the tile images. By increasing the number of tile images, the tolerance for some errors is increased, thereby improving the processing accuracy of the entire deep learning algorithm model. In addition, before the deep learning algorithm model performs the deconvolution operation, multiple corresponding multi-dimensional tile images are fused to enhance the features in the tile images. By enhancing the features of parking spaces or columns through fusion, the accuracy of model processing is improved when processing the deep learning algorithm model.

[0059] Specifically, after the deep learning algorithm model has been processed as described above, the enhanced feature map is expanded, and the parking spaces in the tile map are classified and processed separately. Finally, regression correction is performed on the parking spaces or the detection frames within the detection frame to obtain the corresponding category information and contour information of the dataset, and finally obtain the polygonal outline including the parking space outline and the pillar outline.

[0060] Figure 2 The following is a schematic diagram of the operation flow of an example of the deep learning instance segmentation algorithm model of this application. Figure 2 The process of processing the dataset using the deep learning instance segmentation algorithm model in this application is described as follows:

[0061] First, a dataset consisting of point cloud tiles is input into the deep learning instance segmentation algorithm model. Then, a data augmentation module is used to perform diversified amplification on each point cloud tile in the dataset, including angle rotation, mirror flipping, and random cropping, to improve the generalization ability of the deep learning instance segmentation algorithm. A clustering dimensionality reduction algorithm, such as PCA, kmeans, or NMS, is used to generate a detection box with a higher probability corresponding to each point cloud tile in the dataset, that is, a detection box that is more suitable for the current point cloud tile. A downsampling convolution operation is performed to extract the feature map from each point cloud tile in the dataset. By fusing feature maps of different sizes and dimensions, multi-dimensional features are enhanced to obtain an optimized feature map. The corresponding regional features are then obtained through the detection box. The optimized feature map in each point cloud tile is expanded to the original image size through a deconvolution operation. The optimized feature map and pixel points are then classified. A regression correction calculation is then performed to obtain the corresponding category information, distinguish between parking spaces and pillars, and the corresponding contour information of parking spaces and pillars.

[0062] exist Figure 1 In an implementation manner, the method of extracting coordinates of underground parking lot elements of the present application includes process S104, fusing multiple tile images to obtain a fused image, and in the fused image, determining the longitude and latitude coordinates corresponding to the polygonal outline of the parking lot element based on the relative position of the polygonal outline of the parking lot element and the reference point and the longitude and latitude coordinates of the reference point.

[0063] In this implementation, after processing by a deep learning model, the parking spaces and pillars in the tile image are determined. The tiles are then fused, combining the smaller tiles into a fused image representing the entire underground parking lot. The longitude and latitude coordinates of the parking space and pillar outlines in the fused image are then derived based on the pixel coordinates corresponding to the parking space and pillar outlines, the pixel coordinates of the reference points, and the awe coordinates of the pixel points.

[0064] Optionally, in the fused image, the longitude and latitude coordinates corresponding to the outline of the parking lot element are determined based on the relative positions of the polygonal outline of the parking lot element and the reference point and the longitude and latitude coordinates of the reference point, including: in the fused point image, obtaining the pixel coordinates of the outline of the parking lot element and the pixel coordinates of the reference point, and then determining the relative position of the outline of the parking lot element and the reference point; converting the longitude and latitude coordinates of the reference point according to the relative position to obtain the longitude and latitude coordinates corresponding to the polygonal outline of the parking lot element.

[0065] In this optional embodiment, the pixel coordinates of the reference point, as well as the pixel coordinates of the parking space outline and the pillar outline, can be obtained from the fused image. The relative positions of the parking space outline and the pillar outline relative to the reference point can be determined. Given the longitude and latitude coordinates of the reference point, coordinate conversion can be performed to obtain the corresponding longitude and latitude coordinates of the parking space outline and the pillar outline.

[0066] Optionally, in the fused image, the longitude and latitude coordinates corresponding to the outline of the parking lot element are determined based on the relative position of the outline of the parking lot element and the reference point and the longitude and latitude coordinates of the reference point, and the method also includes: performing a vectorization operation on the polygonal outline of the parking lot element in the fused image, correcting the polygonal outline, and obtaining a rectangular outline of the parking lot element; obtaining the pixel coordinates of the rectangular outline and the pixel coordinates of the reference point, and then determining the relative position of the rectangular outline and the reference point.

[0067] In this embodiment, the acquired parking space and pillar outline information in the underground parking lot is first vectorized and corrected to avoid issues such as offset or missing curves. Coordinate conversion is then performed to convert the pixel coordinates of the parking spaces and pillars into actual latitude and longitude coordinates, facilitating the subsequent creation of underground parking lot maps.

[0068] Specifically, after processing by the deep learning instance segmentation algorithm model, the outline coordinates of each parking space and column in the entire parking lot map have been obtained. However, the outline coordinates are irregular polygonal curve outlines, and some of them are overlapped, missing, offset, etc., so vectorization operations are required to further correct the outline coordinates of the parking spaces and columns and correct the polygonal curve outlines to rectangular outlines.

[0069] Figure 3 A schematic diagram of the process of vectorization operation in the method of extracting parking spaces and column coordinates in an underground parking lot of the present application is shown.

[0070] like Figure 3As shown in the figure, the vectorization process includes area filtering, minimum rectangularization, contour rotation, contour fusion, contour collinearity, and contour padding, ultimately outputting the rectangular outlines of parking spaces and pillars. Through these operations, the outlines of parking spaces and pillars that do not meet the requirements are filtered out, and those with defects such as missing or offset parking spaces and pillars are corrected, ultimately obtaining a more complete outline of the parking spaces and pillars.

[0071] Specifically, because the obtained contour point coordinates of parking spaces and columns are pixel coordinates, not actual longitude and latitude coordinates, further GIS coordinate transformation is required. In the specific coordinate conversion process, a reference point coordinate is saved in each parking lot point cloud map. This reference point has GPS information. According to the distance of the contour coordinates of the parking spaces and columns relative to this reference coordinate, the contour point coordinates of the parking spaces and columns can be converted into actual GPS coordinates, that is, longitude and latitude coordinates. That is, the pixel coordinates of the parking spaces and columns are converted into actual longitude and latitude coordinates. Then, the subsequent production of underground parking maps is carried out, which speeds up the production and reduces labor costs.

[0072] Optionally, the contour information is converted to obtain the coordinates corresponding to the parking space and the pillar, including: splicing and converting the parking space coordinates and the pillar coordinates corresponding to the point cloud tile image in the contour information to obtain the corresponding parking space contour coordinates and the pillar contour coordinates in the point cloud image.

[0073] In this embodiment, when slicing the large parking lot point cloud image, to ensure the integrity of the parking spaces within the point cloud tiles, a preset overlap ratio is maintained between the point cloud tiles and their adjacent point cloud tiles. When the individual point cloud tiles are stitched together to form the large parking lot image, to prevent overlapping portions from being recognized multiple times, the parking space outlines are further filtered and fused, resulting in a single optimally recognized outline for each parking space and pillar.

[0074] Figure 4 A flow chart showing an example of point cloud tile cutting and splicing in the method for obtaining coordinates of underground parking elements in the present application is shown.

[0075] like Figure 4As shown, the large image of the parking lot is cut to obtain multiple point cloud tile images. Among them, in order to ensure the integrity of the parking spaces in the point cloud tile image, each adjacent point cloud tile image maintains a certain value of overlap, such as 1 / 3. The point cloud tile image is encoded. Subsequently, the deep learning instance segmentation algorithm is processed to obtain the contour coordinates of the columns and parking spaces of each point cloud tile image. When the point cloud tile images are spliced ​​into a large image of the parking lot, they are spliced ​​according to the encoding in each point cloud tile image, and contour filtering and fusion operations are performed to avoid the overlap of parking spaces and contours, so that each parking space and column only obtains one optimal recognition contour. Then, when generating the large image of the parking lot, the corresponding parking space and column contour coordinates are obtained. Subsequently, the coordinates of the parking spaces and contours are vectorized and converted to obtain the latitude and longitude coordinates of the parking spaces and columns in the final underground parking lot.

[0076] This application's method for obtaining coordinates of underground parking lot elements uses LiDAR to acquire parking lot point cloud data, obtaining image information and coordinate information for parking spaces and pillars to improve recognition results. A deep learning instance segmentation algorithm model is used to identify parking spaces and pillars in the parking lot. Then, through vectorization and integration with GPS latitude and longitude information, the outline of each parking space and pillar in the parking lot is converted into longitude and latitude coordinates. This method quickly converts parking space and pillar information into map data, significantly reducing the workload of frontline production workers, lowering production costs, and accelerating the production and updating of parking lot maps.

[0077] Figure 5 A specific implementation of the system for extracting coordinates of underground parking elements in the present application is shown.

[0078] exist Figure 5 In the illustrated embodiment, the system for extracting coordinates of underground parking lot elements of the present application includes: a data acquisition module 501, which collects environmental data of the underground parking lot and determines the longitude and latitude coordinates of the reference point in the environmental data; an image conversion and segmentation module 502, which converts the environmental data and segments the obtained initial image to obtain multiple tile images; a contour determination module 503, which marks the parking lot elements in the tile image and processes the tile image using a pre-trained deep learning algorithm model to obtain polygonal contours corresponding to the parking lot elements in each tile image; and a coordinate determination module 504, which fuses multiple tile images to obtain a fused image, and in the fused image, determines the longitude and latitude coordinates corresponding to the polygonal contour of the parking lot element based on the relative position of the polygonal contour of the parking lot element and the reference point and the longitude and latitude coordinates of the reference point.

[0079] Optionally, in the contour determination module 503, a downsampling convolution operation is performed on the tile image to extract a feature map in the tile image; the feature map is processed by a clustering dimensionality reduction algorithm to obtain a detection frame corresponding to the feature map; a deconvolution operation is performed on the feature map, and a regression process is performed on the detection frame in the feature map after the deconvolution operation to determine the polygonal contour of the parking lot element.

[0080] Optionally, in the contour determination module 503, before performing the downsampling convolution operation on the tile image, the tile image is amplified to obtain a multi-dimensional tile image, and the amplification includes angle rotation; before performing the deconvolution operation on the feature map, multiple corresponding multi-dimensional tile images are fused to enhance the features in the tile image.

[0081] Optionally, in the coordinate determination module 504, the pixel coordinates of the outline of the parking lot element and the pixel coordinates of the reference point are obtained in the fusion point image, and then the relative position of the outline of the parking lot element and the reference point is determined; the latitude and longitude coordinates of the reference point are converted according to the relative position to obtain the latitude and longitude coordinates corresponding to the polygonal outline of the parking lot element.

[0082] Optionally, in the coordinate determination module 504, a vectorization operation is performed on the polygonal outline of the parking lot element in the fused image, and the polygonal outline is corrected to obtain a rectangular outline of the parking lot element; the pixel coordinates of the rectangular outline and the pixel coordinates of the reference point are obtained, and then the relative positions of the rectangular outline and the reference point are determined.

[0083] Optionally, in the conversion and segmentation module 502 , the initial image is segmented according to a preset ratio of overlap between adjacent tile images to obtain a plurality of tile images.

[0084] Optionally, in the conversion and segmentation module 502, under the condition that the environmental data is point cloud data, the point cloud data is selected according to the elevation information of the point cloud data to determine the parking lot point cloud data and the pillar point cloud data; the parking lot point cloud data and the pillar point cloud data are converted respectively to obtain an initial image including the parking lot point cloud map and the pillar point cloud map.

[0085] Optionally, in the conversion and segmentation module 502 , a segmentation range is determined according to the size of the parking spaces in the underground parking lot; and the initial image is segmented according to the segmentation range to obtain a plurality of tile images.

[0086] In this application's system for extracting coordinates of underground parking lot elements, laser radar (LiDAR) acquires parking lot point cloud data, obtaining image information and coordinate information for parking spaces and pillars to refine the recognition results. A deep learning instance segmentation algorithm model is used to identify parking spaces and pillars in the parking lot. Then, through vectorization and integration with GPS longitude and latitude information, the outline of each parking space and pillar in the parking lot is converted into longitude and latitude coordinates. Rapidly converting parking space and pillar information into map data significantly reduces the workload of frontline production workers, lowers production costs, and accelerates the process of creating and updating parking lot maps.

[0087] Figure 6 An embodiment of the underground parking lot map making method of the present application is shown.

[0088] exist Figure 6 In the illustrated embodiment, the underground parking lot map production method of the present application includes process S601, collecting environmental data of the underground parking lot and determining the longitude and latitude coordinates of the reference point in the environmental data; process S602, converting the environmental data and segmenting the obtained initial image to obtain multiple tile images; process S603, marking the parking lot elements in the tile image, and processing the tile image using a pre-trained deep learning algorithm model to obtain polygonal contours corresponding to the parking lot elements in each tile image; process S604, fusing the multiple tile images to obtain a fused image, and in the fused image, determining the longitude and latitude coordinates corresponding to the polygonal contour of the parking lot element based on the relative position of the polygonal contour of the parking lot element and the reference point and the longitude and latitude coordinates of the reference point; and process S605, establishing an underground parking lot map based on the polygonal contour and the longitude and latitude coordinates corresponding to the polygonal contour.

[0089] In this embodiment, the specific process of the parking lot map production method of the present application is described in the above embodiment and will not be repeated here. By processing the point cloud data or image data with GPS information collected from the underground parking lot, the obtained parking space outlines and column outlines are provided with longitude and latitude coordinates. Therefore, when mapping the underground parking lot, the collected data can be spatially associated to accurately draw a large map of the underground parking lot.

[0090] In one embodiment of the present application, a computer-readable storage medium stores computer instructions, wherein the computer instructions are operated to execute the method for extracting coordinates of underground parking elements or the method for creating an underground parking map described in any embodiment. The storage medium may be directly implemented in hardware, in a software module executed by a processor, or in a combination of the two.

[0091] The software modules may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor such that the processor can read information from and write information to the storage medium.

[0092] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic, discrete hardware components, or any combination thereof. A general-purpose processor may be a microprocessor, but in the alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors combined with a DSP core, or any other such configuration. In the alternative, the storage medium may be integral to the processor. The processor and storage medium may reside in an ASIC. The ASIC may reside in a user terminal. In the alternative, the processor and storage medium may reside as discrete components in the user terminal.

[0093] In a specific embodiment of the present application, a computer device includes a processor and a memory, the memory storing computer instructions, wherein: the processor operates the computer instructions to execute the method for extracting coordinates of underground parking elements or the method for making an underground parking map described in any embodiment.

[0094] In the embodiments provided in this application, it should be understood that the disclosed devices can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of units is merely a logical function division. In actual implementation, there may be other division methods, such as 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 mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0095] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0096] The above are merely embodiments of the present application and are not intended to limit the patent scope of the present application. Any equivalent structural transformations made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for extracting coordinates of underground parking elements, characterized in that: include: Collecting environmental data of the underground parking lot and determining the latitude and longitude coordinates of reference points in the environmental data; Converting the environmental data and segmenting the converted initial image to obtain a plurality of tile images; Marking the parking lot elements in the tile image, and processing the tile image using a pre-trained deep learning algorithm model to obtain polygonal outlines corresponding to the parking lot elements in each tile image; as well as The processed multiple tile images are fused to obtain a fused image, and in the fused image, the longitude and latitude coordinates corresponding to the polygonal outline of the parking lot element are determined according to the relative position of the polygonal outline of the parking lot element and the reference point and the longitude and latitude coordinates of the reference point.

2. The method for extracting coordinates of underground parking elements according to claim 1, characterized in that: The method of processing the tile images using a pre-trained deep learning algorithm model to obtain polygonal outlines corresponding to the parking lot elements in each tile image includes: Performing a downsampling convolution operation on the tile image to extract a feature map from the tile image; Processing the feature map by a clustering dimensionality reduction algorithm to obtain a detection frame corresponding to the feature map; A deconvolution operation is performed on the feature map, and regression processing is performed on the detection box in the feature map after the deconvolution operation to determine the polygonal outline of the parking lot element.

3. The method for extracting coordinates of underground parking elements according to claim 2, characterized in that: The method further includes processing the tile images using a pre-trained deep learning algorithm model to obtain polygonal outlines corresponding to the parking lot elements in each tile image. Before performing a downsampling convolution operation on the tile image, amplifying the tile image to obtain a multi-dimensional tile image, wherein the amplification includes angular rotation; Before performing the deconvolution operation on the feature map, a plurality of mutually corresponding multi-dimensional tile maps are fused to enhance the features in the tile maps.

4. The method for extracting coordinates of underground parking elements according to claim 1, characterized in that: Determining, in the fused image, the longitude and latitude coordinates corresponding to the outline of the parking lot element based on the relative positions of the polygonal outline of the parking lot element and the reference point and the longitude and latitude coordinates of the reference point includes: In the fusion point image, the pixel coordinates of the outline of the parking lot element and the pixel coordinates of the reference point are obtained, thereby determining the relative position of the outline of the parking lot element and the reference point; The longitude and latitude coordinates of the reference point are converted according to the relative position to obtain the longitude and latitude coordinates corresponding to the polygonal outline of the parking lot element.

5. The method for extracting coordinates of underground parking elements according to claim 4, characterized in that: The method further includes determining the longitude and latitude coordinates corresponding to the outline of the parking lot element according to the relative positions of the outline of the parking lot element and the reference point and the longitude and latitude coordinates of the reference point in the fused image, further comprising: performing a vectorization operation on the polygonal outline of the parking lot element in the fused image, and correcting the polygonal outline to obtain a rectangular outline of the parking lot element; The pixel coordinates of the rectangular outline and the pixel coordinates of the reference point are obtained, and then the relative positions of the rectangular outline and the reference point are determined.

6. The method for extracting coordinates of underground parking elements according to claim 1, characterized in that: The initial image is segmented to obtain a plurality of tile images, including: The initial image is segmented according to a preset ratio of overlap between adjacent tile images to obtain a plurality of tile images.

7. The method for extracting coordinates of underground parking elements according to claim 1, characterized in that: The converting of the environmental data includes: Under the condition that the environmental data is point cloud data, the point cloud data is selected according to elevation information of the point cloud data to determine parking spot point cloud data and column point cloud data; The parking spot point cloud data and the pillar point cloud data are converted respectively to obtain the initial image including the parking spot point cloud map and the pillar point cloud map.

8. The method for extracting coordinates of underground parking elements according to claim 1, characterized in that: The initial image is segmented to obtain a plurality of tile images, including: Determine the segmentation range according to the size of the parking spaces in the underground parking lot; The initial image is segmented according to the segmentation range to obtain a plurality of tile images.

9. A system for extracting coordinates of underground parking elements, characterized in that: include: A data acquisition module, which collects environmental data of the underground parking lot and determines the latitude and longitude coordinates of reference points in the environmental data; An image conversion and segmentation module, which converts the environmental data and segments the obtained initial image to obtain multiple tile images; a contour determination module, which marks the parking lot elements in the tile image and processes the tile image using a pre-trained deep learning algorithm model to obtain polygonal contours corresponding to the parking lot elements in each tile image; as well as A coordinate determination module is configured to fuse the processed multiple tile images to obtain a fused image, and in the fused image, determine the longitude and latitude coordinates corresponding to the polygonal outline of the parking lot element based on the relative position of the polygonal outline of the parking lot element and the reference point and the longitude and latitude coordinates of the reference point.

10. A method for making an underground parking lot map, characterized in that: include: Collecting environmental data of the underground parking lot and determining the latitude and longitude coordinates of reference points in the environmental data; Converting the environmental data and segmenting the obtained initial image to obtain a plurality of tile images; Marking the parking lot elements in the tile image, and processing the tile image using a pre-trained deep learning algorithm model to obtain polygonal outlines corresponding to the parking lot elements in each tile image; Fusing the processed plurality of tile images to obtain a fused image, and determining, in the fused image, the latitude and longitude coordinates corresponding to the polygonal outline of the parking lot element based on the relative position of the polygonal outline of the parking lot element and the reference point and the latitude and longitude coordinates of the reference point; An underground parking lot map is established based on the polygonal outline and the longitude and latitude coordinates corresponding to the polygonal outline.

11. A computer-readable storage medium storing computer instructions, wherein the computer instructions are operated to execute the method for extracting coordinates of underground parking elements according to any one of claims 1 to 8 or the method for producing an underground parking map according to claim 10.

Images