A high-precision mapping and positioning method for cars in underground parking lots

Abstract

This invention discloses a high-precision mapping and positioning method for vehicles in underground parking lots. The method involves mapping first, then positioning. First, a three-dimensional map of the underground parking lot is constructed based on wireless signals and machine vision, comprising a navigation layer, a feature layer, and a structural layer. Then, based on this 3D map, mathematical models and various matching algorithms are used to match the vehicle's current scene with features from the 3D map, achieving high-precision positioning of the vehicle, especially in large underground parking lots, even in completely enclosed areas where satellite positioning signals are weak or lost. This invention deeply integrates wireless signals and visual features, eliminating the need for signal base stations and relying on maps provided by other operators. High-precision positioning in underground parking lots can be achieved solely through the vehicle's built-in monocular camera and wireless signal receiver, combining economy and accuracy, with simple operation and easy promotion.

Description

Technical Field

[0001] This invention belongs to the field of smart parking, and in particular relates to a method for high-precision mapping and positioning of vehicles in underground parking lots. Background Technology

[0002] Currently, in the field of smart parking in open traffic scenarios, cars can achieve good positioning results with the assistance of differential GPS / BeiDou combined with inertial navigation.

[0003] However, in some complex traffic scenarios, such as overpasses, dense clusters of tall buildings, and underground parking lots, the relatively enclosed environment can lead to the loss of GPS satellites, causing drift errors. Especially in underground parking lots, the completely enclosed area creates GPS blind spots, and high-precision inertial navigation systems like NovAtel SPAN-IGM-S1 will fail within 20 seconds. Therefore, positioning in such enclosed scenarios is one of the most pressing challenges for high-precision positioning. Summary of the Invention

[0004] Based on the problems mentioned above, the purpose of this invention is to provide a high-precision vehicle positioning and navigation method in underground parking lots, which does not require relying on maps provided by other operators or establishing signal base stations, and can achieve high-precision vehicle positioning in completely enclosed areas where satellite positioning signals are weak or lost, especially in large underground parking lots.

[0005] The technical solution to achieve the purpose of this invention is as follows:

[0006] A method for high-precision mapping and positioning of vehicles in an underground parking lot includes the following steps:

[0007] Step 1: Collect wireless signals and scene features of the underground parking lot to form a navigation layer, feature layer, and structure layer, and construct a 3D map of the underground parking lot;

[0008] Step 2: Based on the 3D map of the underground parking lot constructed in Step 1, using the car's wireless signal receiver and monocular camera, a BoW model is established to match the scene of the vehicle's location captured by the monocular camera with the scene features in the constructed 3D map, thereby determining the parameters of the positioning matrix and vector, and thus determining the precise location of the car.

[0009] Compared with the prior art, the significant advantages of this invention are:

[0010] (1) The technical solution of the present invention uses visual and wireless signals to complete positioning, which solves the problem of no GPS / BeiDou signal in the underground parking lot environment and has universality for closed space scenarios where satellite signals are weak or lost;

[0011] (2) In the process of mapping and positioning, the technical solution of the present invention fully considers economic cost factors while ensuring positioning accuracy. The mapping process uses binocular vision and wireless signals to construct the built-in scene of the underground parking lot through mathematical modeling, without the need to establish a signal base station. The navigation process only needs to collect the current location scene features through the monocular camera equipped in the vehicle and use various feature algorithms to perform feature matching to achieve high-precision positioning of the vehicle in the underground parking lot.

[0012] (3) The technical solution of the present invention takes into account both accuracy and economy, and the required equipment is simple and the process is easy to operate.

[0013] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating the steps involved in constructing a three-dimensional map in an embodiment of the present invention.

[0015] Figure 2 This is a flowchart illustrating the steps for high-precision positioning in an embodiment of the present invention.

[0016] Figure 3 This is a schematic diagram comparing the error between the constructed 3D map and the actual map in an embodiment of the present invention.

[0017] Figure 4 This is a schematic diagram comparing the actual error of high-precision positioning in an embodiment of the present invention. Detailed Implementation

[0018] A method for high-precision mapping and positioning of vehicles in an underground parking lot includes the following steps:

[0019] Step 1: Collect wireless signals and scene features from the underground parking lot to form a navigation layer, feature layer, and structure layer, and construct a 3D map of the underground parking lot. Specifically:

[0020] Step 1-1: Deploy N data collection points in the underground parking lot and select any one of them;

[0021] Steps 1-2: Use a wireless signal receiver and a binocular camera to collect wireless signal data and scene images at the selected collection points. The collected wireless signals include, but are not limited to, WIFI, UWB, Bluetooth, and other data.

[0022] Steps 1-3: Use the wireless signal data collected in Step 1-2 to form the navigation layer of the 3D map of the underground parking lot. That is, by storing the wireless signal strength of the point, a navigation layer is formed in the construction of the 3D map, which provides a reference for the initial coarse positioning of vehicles in the underground parking lot.

[0023] Steps 1-4: Extract image features from the scene images acquired in Step 1-2 to form the feature layer of the 3D map of the underground parking lot. Specifically:

[0024] A BoW model is constructed from the collected scene images, and algorithms such as, but not limited to, ORB, SIFT, and SURF are used to extract global features, local features, and BoW global features from the scene images, thus obtaining all features of the scene images:

[0025]

[0026] in, Let j represent the feature set of the j-th data collection point. This represents the nth feature in the set.

[0027] Steps 1-5: Based on the extracted image features, use Hamming distance calculation to perform feature matching on the left and right scenes of the binocular camera, and use, but not limited to, triangulation to perform 3D reconstruction on the matched feature points, so that each feature point of the feature matching has 3D information, forming the structural layer of the 3D map of the underground parking lot, and completing the wireless signal and scene image processing work of the data collection point.

[0028] Steps 1-6: Repeat steps 1-2 to 1-5 for the next data collection point until the data collection and processing of N data collection points in the underground parking lot are completed, and the 3D map of the underground parking lot is constructed.

[0029] Step 2: Based on the 3D map of the underground parking lot constructed in Step 1, using the car's wireless signal receiver and monocular camera, a BoW model is established. The scene of the vehicle's location captured by the monocular camera is matched with the scene features in the constructed 3D map to determine the parameters of the positioning matrix and vectors, thereby determining the precise location of the car. Specifically:

[0030] Step 2-1: Drive the car equipped with a wireless signal receiver and a monocular camera to any location in the underground parking lot. Use this location as the location to be located. Use the wireless signal receiver and monocular camera to collect wireless signal data and scene images of the location to be located. The collected wireless signals include, but are not limited to, WIFI, UWB, Bluetooth, etc.

[0031] Step 2-2: Extract image features from the scene images acquired in Step 2-1, specifically as follows:

[0032] A BoW model is constructed from the collected scene images, and algorithms such as, but not limited to, ORB, SIFT, and SURF are used to extract global features, local features, and BoW global features from the scene images, thus obtaining all features of the scene images:

[0033]

[0034] in, This represents the feature set of the point to be located. This represents the m-th feature in the set.

[0035] Step 2-3: Based on the navigation layer of the 3D map constructed in Step 1 and the wireless signal data collected in Step 2-1, perform coarse vehicle positioning, specifically as follows:

[0036] Construct a Gaussian weighted AP model to match the wireless signal data collected by the car with the wireless signal data of each data collection point in the 3D map of the underground parking lot:

[0037]

[0038] Where n is the number of MPC matches, and L and M are the RSSI values;

[0039] Based on the matching results, the data collection point closest to the vehicle is obtained. Using this data collection point as the center, four data collection points are selected in front of, behind, to the left and right of it.

[0040] Steps 2-4: Based on the feature layer of the 3D map constructed in Step 1, match it with the coarse localization results from Step 2-3 to further locate the vehicle. Specifically:

[0041] Step 2-4-1: Based on the 5 data collection points obtained in Step 2-3, determine the Euclidean distance between the global features, local features, and BoW global features of the image features of the current location point of the car and the image features of the data collection points:

[0042]

[0043] Where X is the feature set of the point to be located, Y is the feature set of the 5 data collection points, and x i With y i Let i be the i-th global feature point, local feature point, or BoW global feature point in the feature set, and k represent the number of global feature points, local feature points, and BoW global feature points, respectively.

[0044] Step 2-4-2: Determine a data acquisition point for each of the global feature points, local feature points, and BoW global feature points;

[0045] Global feature points: Based on the minimum Euclidean distance between the global features of the current location point of the vehicle and the data collection points in the image features, determine the data collection point that is closest to the location point among the 5 data collection points;

[0046] Local feature points: Based on the set Euclidean distance threshold, when the local features of each data collection point are less than the threshold, they are considered matching points. Finally, by comparing the number of matching points of each data collection point, the data collection point with the most matching points is selected as the data collection point closest to the car.

[0047] BoW Global Feature Points: By comparing the Euclidean distances between each data collection point and the BoW global feature points of the point to be located, the data collection point with the highest probability is determined to be the data collection point closest to the car.

[0048] Step 2-4-3: Based on the three data collection points obtained in Step 2-4-2, the K-nearest neighbor model is used to finally determine the data collection point closest to the car, that is, to determine the actual position of the car near the data collection point, thus completing the further positioning of the point to be located.

[0049] Steps 2-5: Based on the structural layers of the 3D map constructed in Step 1 and the positioning results in Step 2-4, the precise positioning of the vehicle is completed, specifically as follows:

[0050] Based on the structural layers of the 3D map of the underground parking lot in step 1 and the data collection point closest to the point to be located determined in steps 2-4, a perspective N-point problem model, namely the PnP model, is constructed:

[0051]

[0052] Where K represents the intrinsic parameter matrix of the car monocular camera, R is the rotation matrix to be determined, and t is the translation vector to be determined. It is the i-th local feature point of the current location to be located of the car. It is the 3D point corresponding to the i-th local feature of the scene at the defined data acquisition point;

[0053] Based on the above formula, the rotation matrix R and translation vector t are determined, thereby determining the offset between the current position of the car and the determined data acquisition point, thus determining the precise position of the car and completing the precise positioning of the car.

[0054] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0055] Example

[0056] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0057] A method for high-precision mapping and positioning of vehicles in an underground parking lot includes the following steps:

[0058] Step 1, Combining Figure 1 The system collects wireless signals and scene features from underground parking lots to form a navigation layer, a feature layer, and a structure layer, thereby constructing a 3D map of the underground parking lot. Specifically:

[0059] Step 1-1: Deploy N data collection points in the underground parking lot and select any one of them;

[0060] Steps 1-2: Use a wireless signal receiver and a binocular camera to collect wireless signal data and scene images at the selected collection points. The collected wireless signals include, but are not limited to, WIFI, UWB, Bluetooth, and other data.

[0061] Steps 1-3: Use the wireless signal data collected in Step 1-2 to form the navigation layer of the 3D map of the underground parking lot. That is, by storing the wireless signal strength of the point, a navigation layer is formed in the construction of the 3D map, which provides a reference for the initial coarse positioning of vehicles in the underground parking lot.

[0062] Steps 1-4: Extract image features from the scene images acquired in Step 1-2 to form the feature layer of the 3D map of the underground parking lot. Specifically:

[0063] A BoW model is constructed from the collected scene images, and algorithms such as, but not limited to, ORB, SIFT, and SURF are used to extract global features, local features, and BoW global features from the scene images, thus obtaining all features of the scene images:

[0064]

[0065] in, Let j represent the feature set of the j-th data collection point. This represents the nth feature in the set.

[0066] Steps 1-5: Based on the extracted image features, use Hamming distance calculation to perform feature matching on the left and right scenes of the binocular camera, and use, but not limited to, triangulation to perform 3D reconstruction on the matched feature points, so that each feature point of the feature matching has 3D information, forming the structural layer of the 3D map of the underground parking lot, and completing the wireless signal and scene image processing work of the data collection point.

[0067] Steps 1-6: Repeat steps 1-2 to 1-5 for the next data collection point until the data collection and processing of N data collection points in the underground parking lot are completed, and the 3D map of the underground parking lot is constructed.

[0068] Step 2, Combining Figure 2 Based on the 3D map of the underground parking lot constructed in step 1, using the car's wireless signal receiver and monocular camera, a BoW model is established. The scene of the vehicle's location captured by the monocular camera is matched with the scene features in the constructed 3D map to determine the parameters of the positioning matrix and vector, thereby determining the precise location of the car. Specifically:

[0069] Step 2-1: Drive the car equipped with a wireless signal receiver and a monocular camera to any location in the underground parking lot. Use this location as the location to be located. Use the wireless signal receiver and monocular camera to collect wireless signal data and scene images of the location to be located. The collected wireless signals include, but are not limited to, WIFI, UWB, Bluetooth, etc.

[0070] Step 2-2: Extract image features from the scene images acquired in Step 2-1, specifically as follows:

[0071] A BoW model is constructed from the collected scene images, and algorithms such as, but not limited to, ORB, SIFT, and SURF are used to extract global features, local features, and BoW global features from the scene images, thus obtaining all features of the scene images:

[0072]

[0073] in, This represents the feature set of the point to be located. This represents the m-th feature in the set.

[0074] Step 2-3: Based on the navigation layer of the 3D map constructed in Step 1 and the wireless signal data collected in Step 2-1, perform coarse vehicle positioning, specifically as follows:

[0075] Construct a Gaussian weighted AP model to match the wireless signal data collected by the car with the wireless signal data of each data collection point in the 3D map of the underground parking lot:

[0076]

[0077] Where n is the number of MPC matches, and L and M are the RSSI values;

[0078] Based on the matching results, the data collection point closest to the vehicle is obtained. Using this data collection point as the center, four data collection points are selected in front of, behind, to the left and right of it.

[0079] Steps 2-4: Based on the feature layer of the 3D map constructed in Step 1, match it with the coarse localization results from Step 2-3 to further locate the vehicle. Specifically:

[0080] Step 2-4-1: Based on the 5 data collection points obtained in Step 2-3, determine the Euclidean distance between the global features, local features, or BoW global features of the image features of the current location point of the car and the image features of the data collection points:

[0081]

[0082] Where X is the feature set of the point to be located, Y is the feature set of the 5 data collection points, and x i With y i Let i be the i-th global feature point, local feature point, or BoW global feature point in the feature set, and k represent the number of global feature points, local feature points, or BoW global feature points, respectively.

[0083] Step 2-4-2: Determine a data acquisition point for each of the global feature points, local feature points, and BoW global feature points;

[0084] Global feature points: Based on the minimum Euclidean distance between the global features of the current location point of the vehicle and the data collection points in the image features, determine the data collection point that is closest to the location point among the 5 data collection points;

[0085] Local feature points: Based on the set Euclidean distance threshold, when the local features of each data collection point are less than the threshold, they are considered matching points. Finally, by comparing the number of matching points of each data collection point, the data collection point with the most matching points is selected as the data collection point closest to the car.

[0086] BoW Global Feature Points: By comparing the Euclidean distances between each data collection point and the BoW global feature points of the point to be located, the data collection point with the highest probability is determined to be the data collection point closest to the car.

[0087] Step 2-4-3: Based on the three data collection points obtained in Step 2-4-2, the K-nearest neighbor model is used to finally determine the data collection point closest to the car, that is, to determine the actual position of the car near the data collection point, thus completing the further positioning of the point to be located.

[0088] Steps 2-5: Based on the structural layers of the 3D map constructed in Step 1 and the positioning results in Step 2-4, the precise positioning of the vehicle is completed, specifically as follows:

[0089] Based on the structural layers of the 3D map of the underground parking lot in step 1 and the data collection point closest to the point to be located determined in steps 2-4, a perspective N-point problem model, namely the PnP model, is constructed:

[0090]

[0091] Where K represents the intrinsic parameter matrix of the car monocular camera, R is the rotation matrix to be determined, and t is the translation vector to be determined. It is the i-th local feature point of the current location to be located of the car. It is the 3D point corresponding to the i-th local feature of the scene at the defined data acquisition point;

[0092] Based on the above formula, the rotation matrix R and translation vector t are determined, thereby determining the offset between the current position of the car and the determined data acquisition point, thus determining the precise position of the car and completing the precise positioning of the car.

[0093] Combination Figure 3 ,pass Figure 1 The provided method for constructing 3D maps of underground parking lots is used as an example to conduct experimental verification of the accuracy of the method:

[0094] 100 data collection points were randomly selected in the enclosed space of an actual underground parking lot. Following the steps in step 1, the three-dimensional map construction process of the actual underground parking lot was completed. The average error of the accuracy of a single data collection point was 4.57 cm, and the overall standard deviation was 3.62 cm.

[0095] Combination Figure 4 ,pass Figure 2 The high-precision positioning method based on the above-mentioned construction of a 3D map of an underground parking lot is used as an example to conduct an accuracy experiment to verify the high-precision positioning method based on the 3D map of an underground parking lot:

[0096] One hundred data collection points were randomly selected in the enclosed space of an actual underground parking lot. Following the operation in step 2, a high-precision positioning experiment of the car in the actual underground parking lot was completed. The average error of the positioning accuracy of a single data collection point was 7.96cm, and the overall standard deviation was 6.13cm.

[0097] In summary, the technical solution of this invention uses only binocular vision and wireless signals to realize mapping and positioning throughout the entire process. It constructs the built-in scene of the underground parking lot through mathematical modeling, without the need to establish signal base stations. The navigation process only needs to collect the current location scene features through the monocular camera equipped in the vehicle and use various feature algorithms for feature matching to achieve high-precision positioning of the vehicle in the underground parking lot. The equipment is simple and the process is easy to operate.

Claims

1. A method for high-precision mapping and positioning of vehicles in an underground parking lot, characterized in that, Includes the following steps: Step 1: Collect wireless signals and scene features of the underground parking lot to form a navigation layer, feature layer, and structure layer, and construct a 3D map of the underground parking lot; Step 2: Based on the 3D map of the underground parking lot constructed in Step 1, using the car's wireless signal receiver and monocular camera, a BoW model is established. The scene of the vehicle's location captured by the monocular camera is matched with the scene features in the constructed 3D map to determine the parameters of the positioning matrix and vectors, thereby determining the precise location of the car. Step 2-1: Drive the car equipped with a wireless signal receiver and a monocular camera to the location to be located in the underground parking lot, and use the wireless signal receiver and monocular camera to collect wireless signal data and scene images of the location to be located. Step 2-2: Extract image features from the scene images acquired in Step 2-1; Step 2-3: Based on the navigation layer of the 3D map constructed in Step 1 and the wireless signal data collected in Step 2-1, perform coarse vehicle positioning: Construct a Gaussian weighted AP model to match the wireless signal data collected by the car with the wireless signal data of each data collection point in the 3D map of the underground parking lot: ； Where n is the number of MPC matches, and L and M are the RSSI values; Based on the matching results, the data collection point closest to the vehicle is obtained. Using this data collection point as the center, four data collection points are selected in front of, behind, to the left and right of it. Steps 2-4: Match the feature layers of the 3D map constructed in Step 1 with the coarse localization results from Step 2-3 to further locate the vehicle. Step 2-4-1: Based on the 5 data collection points obtained in Step 2-3, determine the Euclidean distance between the global features, local features, and BoW global features of the image features of the current location point of the car and the image features of the data collection points: ； Where X is the feature set of the point to be located, Y is the feature set of the 5 data collection points, and x i With y i Let i be the i-th global feature point, local feature point, and BoW global feature point in the feature set, respectively, and k represent the number of global feature points, local feature points, and BoW global feature points, respectively. Step 2-4-2: Determine a data acquisition point for each of the global feature points, local feature points, and BoW global feature points; Global feature points: Based on the minimum Euclidean distance between the global features of the current location point of the vehicle and the data collection points in the image features, determine the data collection point that is closest to the location point among the 5 data collection points; Local feature points: Based on the set Euclidean distance threshold, when the local features of each data collection point are less than the threshold, they are considered matching points. Finally, by comparing the number of matching points of each data collection point, the data collection point with the most matching points is selected as the data collection point closest to the car. BoW Global Feature Points: By comparing the Euclidean distances between each data collection point and the BoW global feature points of the point to be located, the data collection point with the highest probability is determined to be the data collection point closest to the car. Step 2-4-3: Based on the three data collection points obtained in Step 2-4-2, the K-nearest neighbor model is used to finally determine the data collection point closest to the car, and the actual position of the car is determined to be near this data collection point, thus completing the further localization of the point to be located. Steps 2-5: Based on the structural layers of the 3D map constructed in Step 1 and the positioning results in Step 2-4, the precise positioning of the car is completed.

2. The high-precision mapping and positioning method for vehicles in underground parking lots according to claim 1, characterized in that, The construction of the 3D map of the underground parking lot in step 1 specifically involves: Step 1-1: Deploy N data collection points in the underground parking lot and select any one of them; Steps 1-2: Use a wireless signal receiver and a binocular camera to collect wireless signal data and scene images at the selected collection points; Steps 1-3: Use the wireless signal data collected in Step 1-2 to form the navigation layer of the three-dimensional map of the underground parking lot; Steps 1-4: Extract image features from the scene images collected in Step 1-2 to form the feature layer of the 3D map of the underground parking lot; Steps 1-5: Based on the extracted image features, perform feature matching on the left and right scenes of the binocular camera, and perform 3D reconstruction on the matched feature points so that each feature point has 3D information, forming the structural layer of the 3D map of the underground parking lot. Steps 1-6: Repeat steps 1-2 to 1-5 for the next data collection point until the data collection and processing of N data collection points in the underground parking lot are completed, and the 3D map of the underground parking lot is constructed.

3. The high-precision mapping and positioning method for vehicles in underground parking lots according to claim 2, characterized in that, The extraction of image features from the scene images of the data collection points in steps 1-4 specifically involves: A BoW model is constructed from the collected scene images, and global features, local features, and BoW global features are extracted from the scene images to obtain all the features of the scene images: ； in, Let j represent the feature set of the j-th data collection point. This represents the nth feature in the set.

4. The high-precision mapping and positioning method for vehicles in underground parking lots according to claim 1, characterized in that, The extraction of image features from the scene image of the point to be located in step 2-2 is specifically as follows: A BoW model is constructed from the collected scene images, and global features, local features, and BoW global features are extracted from the scene images to obtain all the features of the scene images: ； in, This represents the feature set of the point to be located. This represents the m-th feature in the set.

Images