Method and device for classifying road intersections

Abstract

The invention relates to a computer-implemented method for classifying road intersections using GPS data, in particular from several vehicles of a vehicle fleet, comprising the following steps: using existing GPS data points (10, 11) of vehicle movements, which were recorded in particular by means of GPS sensors of the vehicles at at least one road intersection, Converting the recorded GPS data points into a 2D contour by determining a center point (13) of the road intersection and creating a polygon (12) which represents a 2D contour of the road intersection, Encoding the 2D contour of the road intersection using Elliptic Fourier Descriptors (EFDs) via a Fourier transformation procedure and classifying the encoded data into different intersection types.

Description

State of the art

[0001] The state of the art includes several methods for mapping roads and detecting intersections using GPS data, sensor data and various algorithms to generate accurate maps and intersection information.

[0002] US20230341239A1 discloses a system for automatically mapping a road segment comprising at least one processor that receives a plurality of images taken by at least one camera mounted on a vehicle while the vehicle traveled the road segment, converts each of the plurality of images into a corresponding top-down image to provide a plurality of top-down images, aggregates the plurality of top-down images to provide an aggregated top-down image of the road segment, analyzes the aggregated top-down image to identify at least one road feature associated with the road segment, automatically annotates the at least one road feature relative to the aggregated top-down image, and outputs the aggregated top-down image, including the annotated at least one road feature, to at least one memory.

[0003] The object of the invention is to provide a method for the precise classification of road junctions based on GPS data. Disclosure of the invention

[0004] The invention relates to a computer-implemented method for classifying road intersections using GPS data from multiple vehicles in a fleet. The method comprises using existing GPS data points of vehicle movements at road intersections and converting this data into a 2D contour by determining the center point of the intersection and creating a polygon. This contour is then encoded using elliptic Fourier descriptors (EFDs) and processed by Fourier transformation to classify the road intersections into different intersection types.

[0005] GPS data points are the positions recorded by vehicles' GPS sensors. These data points provide information about vehicle movements near or within intersections. A 2D contour is a two-dimensional representation of the intersection geometry, created by connecting the vertices of a polygon that encloses the intersection. Elliptic Fourier descriptors are mathematical tools used to encode the shape of the 2D contour, preparing the data for classification. The Fourier transform is a mathematical procedure for decomposing the contour data into frequency components, specifically Elliptic Fourier descriptors, which are relevant for classification.

[0006] One advantage of this method is that the precise coding of intersection geometries and their subsequent classification enables more accurate identification and categorization of intersection types. This leads to improved analysis of road infrastructure and supports applications in navigation and autonomous driving.

[0007] Advantageously, the classification of intersection types can be performed by a trained deep neural network (DNN). A deep neural network is a multi-layered neural network trained on large datasets to recognize complex patterns.

[0008] This further increases the accuracy of the crossing type classification, as the DNN is able to learn from a wide variety of examples and detect subtle differences between different crossing types.

[0009] Advantageously, the DNN can be trained on the intersection types of a T-junction, an X-junction, and / or a roundabout. A T-junction is a road intersection where one road meets another, forming a T-shape. An X-junction is a four-way intersection where two roads intersect. A roundabout is a circular intersection where traffic flows in a circle around a central island.

[0010] This allows the system to be specifically trained to recognize the most common types of intersections, which facilitates its use in real traffic environments.

[0011] Advantageously, the 2D contour of the road intersection can be formed by iteratively calculating the vertices of the polygon around the intersection's center point. An iterative calculation means that the polygon's vertices are determined step by step through repeated calculations to achieve the best possible approximation of the actual intersection geometry.

[0012] This improves the accuracy of the intersection's contour representation, enabling more precise classification.

[0013] Advantageously, the Fourier coefficients or Elliptic Fourier Descriptors (EFDs) can be calculated by a forward transformation of the 2D contour of the road intersection. Fourier coefficients or Elliptic Fourier Descriptors (EFDs) are the parameters that describe the frequency components of the contour data, and a forward transformation is the process of calculating these coefficients.

[0014] This efficiently converts the contour data into a format, such as a numerical matrix, that is suitable for further processing and classification.

[0015] Advantageously, the Fourier coefficients can be encoded into a fixed numerical matrix to ensure a uniform input size for the DNN. A fixed numerical matrix is ​​a structured representation of the data in a fixed format to ensure that the DNN always receives consistent input data.

[0016] This ensures consistent processing of the data in the neural network, which increases classification efficiency.

[0017] Advantageously, the DNN can classify road intersections in real time or near real time. Real-time classification means that the intersections are analyzed and classified directly during data acquisition.

[0018] This enables an immediate and dynamic analysis of the traffic environment, which is particularly advantageous for autonomous driving applications.

[0019] Advantageously, road intersections can be entered into a digital map along with their classified intersection types. A digital map is a graphical representation of the road infrastructure in which the classified intersections are stored and displayed.

[0020] This provides a visualization of the road infrastructure with additional information on the types of intersections, which facilitates navigation in traffic.

[0021] Advantageously, the collected GPS data can be supplemented by sensor data from other sensors such as a video camera, radar sensor, and / or lidar sensor. Additional sensor data provides further information about the traffic environment, which supports the GPS data.

[0022] This improves the precision of intersection detection, as multiple sensor data sources together enable a more comprehensive analysis.

[0023] Advantageously, the classified intersection types can be used to improve a navigation system to assist a driver and / or to improve the control system of a self-driving or semi-autonomous vehicle. Navigation systems use this data to optimize route guidance, while control systems in autonomous vehicles use the intersection types to control the self-driving vehicle.

[0024] This improves both the safety and efficiency of traffic flow by enabling the systems to respond better to specific types of intersections.

[0025] Advantageously, the coded data can be used to continuously improve the DNN by using the newly coded data as training data for the DNN and / or by automatically detecting new intersection types. This means that the DNN is constantly being trained with new data to further improve its classification.

[0026] This continuously increases the system's performance, as it can adapt to new traffic situations and intersection types.

[0027] Advantageously, the GPS data points can be filtered based on predefined criteria such as vehicle speed, direction of travel, and / or distance to the center of the intersection to eliminate erroneous GPS data points. This ensures a more precise representation of the intersection geometry.

[0028] This further increases the accuracy of the data by eliminating noise and faulty GPS data points.

[0029] Advantageously, the vertices of the polygons can be calculated iteratively by weighting the GPS data points according to the vehicle's movement and / or direction of travel. This weighting ensures that the relevant GPS data is given greater consideration when calculating the contours.

[0030] This allows the contour to be adapted even more precisely to the actual movements of the vehicles.

[0031] Advantageously, the Fourier transform can be optimized so that at most the first ten elliptic Fourier descriptors (EFDs) are used for encoding, thus minimizing the amount of data. This reduces the complexity of the data and avoids distortions.

[0032] This increases the efficiency of the system without affecting the accuracy of the classification.

[0033] Advantageously, a defined numerical matrix can be normalized with Elliptic Fourier Descriptors (EFDs) for input into the DNN, thus ensuring better comparability between different crossing types. Normalization means that the data is converted into a standardized format to enable consistent processing.

[0034] This makes the classification of crossbreed types more consistent and reliable.

[0035] A further object of the invention is a device for carrying out the method comprising a processing unit configured to process GPS data points of vehicle movements at the center of the intersection based on the acquired GPS data and subsequently generates a polygon representing the intersection. The 2D contour serves as the basis for further encoding and classification of the intersection. The device also comprises a storage unit configured to store the calculated 2D contours and / or encoded Fourier coefficients. a classification unit that is set up to classify the coded data of the road intersection contour into different intersection types using a trained Deep Neural Network (DNN), and a communication interface that is set up to transmit the classified intersection types to a navigation system and / or a control system of a vehicle.

[0036] The device comprises a processing unit that processes GPS data points from vehicle movements at at least one intersection. These GPS data points are used to create a 2D contour of the intersection. To do this, the processing unit first determines the center point of the intersection based on the acquired GPS data and then generates a polygon representing the intersection. The 2D contour serves as the basis for further coding and classification of the intersection.

[0037] The storage unit is configured to store the 2D contours generated by the processing unit and / or the coded Fourier coefficients calculated from them. This makes the data available for subsequent processing steps and applications.

[0038] The device also includes a classification unit designed to classify the encoded data of the road intersection contour into different intersection types using a trained deep neural network (DNN). The classification unit utilizes the mathematical properties of the encoded contours to reliably identify intersection types such as T-junctions, X-junctions, or roundabouts.

[0039] A communication interface enables the transmission of classified intersection types to a navigation system and / or a vehicle's control system, for example, to an external server for further processing. This transmission ensures that the classified data can be used in real time or near real time in applications such as route planning or the control of autonomous vehicles.

[0040] The device's modular structure allows for flexible integration into existing vehicle and traffic infrastructures. It ensures precise and efficient processing of large data volumes and supports dynamic adaptation to different traffic conditions and intersection types. Brief description of the drawings

[0041] The invention is explained with reference to the following drawings: Fig. 1 a schematic representation of the first 10 harmonic functions of a typical T-shaped intersection contour; Fig. 2 a schematic representation of the paths of the recorded GPS data points and the generated 2D contour; Fig. Figure 3 shows a schematic representation of an embodiment of a device for carrying out the method described above. Examples of implementation

[0042] Fig. Figure 1 schematically shows the first 10 harmonic functions of a typical T-shaped intersection contour generated by applying Elliptic Fourier Descriptors (EFDs) to the GPS data of the vehicle fleet. The underlying algorithm utilizes the properties of the EFDs to represent a closed 2D contour as a finite series of elliptic harmonic functions. This method is similar to the Fourier transform of a one-dimensional signal, where the waveform is approximated by a finite number of sine and cosine terms.

[0043] In this case, each harmonic component of the intersection's contour represents specific geometric features of the intersection. The number of harmonic functions used—here, the first ten, from the first harmonic function 1 to the tenth harmonic function 2—determines the accuracy of the contour representation. The quality of the intersection's contour representation is influenced by the number of harmonics: as the number increases, the accuracy of the contour representation improves. These harmonics are described by four constant parameters for each harmonic: a, bn, cn, and dn. These parameters control the shape of the elliptical components that contribute to the contour reconstruction.

[0044] Empirical studies show that a minimum of 10 harmonics is required for a typical intersection to achieve sufficient accuracy of over 90% in representing the intersection contour. This is particularly important because the quality of the contour significantly contributes to the precise classification of the intersections, which is performed in subsequent steps by a deep neural network (DNN). The theoretical basis for this accuracy is also supported by studies on Fourier transforms.

[0045] In the next stage of the process, the 2D contour is encoded by calculating the Fourier coefficients for the generated polygon. The result of this step is a data structure containing the four parameters an, bn, cn, and dn for each of the n harmonic functions, where n represents the number of harmonic functions used. The Fourier coefficients are then used to encode the contour into a fixed numerical matrix, which is used for training the DNN.

[0046] Fig. Figure 2 schematically shows, in the left-hand diagram, a sketch of paths 10 and 11 of the recorded GPS data points of the individual vehicles, represented as discontinuous lines in a 3D space. These paths 10 and 11 of the GPS data points encompass the recorded vehicle routes at road intersections and represent the starting point for contour construction. Contour construction is the first step of the proposed algorithm, which aims to transform the vehicle paths 10 and 11 into a 2D representation. Elevation data is ignored to reduce complexity, and the data points are converted into a 2D contour that forms polygon 12 in the right-hand diagram.

[0047] The next step in contour construction, as indicated in the right-hand diagram, is to iteratively transform the vehicle paths into polygon 12. First, a center point or centroid 13 is calculated from the GPS data, followed by the creation of a minimal initial polygon containing the three points closest to the center point. The polygon is then iteratively expanded by adding further GPS points until all GPS data points are included in the contour. This method ensures that the intersection contour completely encloses the recorded vehicle paths and accurately represents the intersection geometry.

[0048] In the next phase of the process, the resulting contour is decomposed using elliptic Fourier descriptors. Here, the contour is divided into a series of elliptic harmonic functions such as... Fig. As described in section 1, the number of harmonic functions used determines the accuracy of the contour reconstruction. Empirically, it has been found that at least 10 harmonic functions are required to achieve an accuracy of over 90%. However, in cases where there are no memory or computation time limitations, up to 50 harmonic functions can be used to achieve nearly 100% contour accuracy.

[0049] After the contour is encoded, the Deep Neural Network (DNN) is trained to classify the intersection contours into user-defined classes such as T-junctions, X-junctions, or roundabouts. The specific DNN model can utilize various neural network architectures to solve this classification problem. The classification is based on the previously encoded data, which is generated using Fourier coefficients. Once sufficient data is available, the DNN is trained to accurately classify future input data, enabling robust classification of road intersections.

[0050] Fig. Figure 3 shows a schematic representation of an embodiment of a device for carrying out the method described above. The device consists of four functional units, which are represented as rectangular boxes and connected by arrows that illustrate the data flow between the units.

[0051] The first box represents the processing unit 20, which is configured to process GPS data points 10 and 11 of vehicle movements at at least one road intersection. This unit receives the GPS data points 10 and 11 and converts them into a 2D contour of the road intersection. To do this, a center point of the intersection 13 is first determined, and then a polygon 12 is generated that represents the geometry of the road intersection. The arrow connecting to the storage unit 21 symbolizes the transfer of the processed data.

[0052] The second box represents the storage unit 21, which is configured to store the calculated 2D contours and / or the Fourier coefficients encoded from them. This storage unit serves as a buffer to hold the processed data for subsequent steps. The arrow connecting to the classification unit 22 indicates that the stored data is transferred to the next processing stage.

[0053] The third box symbolizes the classification unit 22, which is set up to analyze the coded data of the road intersection contour. This unit uses a trained deep neural network (DNN) to classify the road intersections into different intersection types. Examples of such intersection types are T-junctions, X-junctions, and roundabouts. The arrow connecting to the communication interface 23 shows the transfer of the classified results.

[0054] The fourth box represents the communication interface 23, which is configured to transmit the classified intersection types, for example, to an external server for further processing, to a navigation system, and / or to a vehicle's control system. This transmission takes place in real time or near real time to make the classified data usable for applications such as navigation or autonomous driving.

[0055] The arrows between the boxes illustrate the sequential process of data processing, from the acquisition and processing of GPS data to the transmission of the classified intersection types to the target systems. QUOTES INCLUDED IN THE DESCRIPTION

[0000] This list of documents cited by the applicant was automatically generated and is included solely for the reader's convenience. The list is not part of the German patent or utility model application. The DPMA accepts no liability for any errors or omissions. Cited patent literature

[0000] US 20230341239A1

[0002]

Claims

[1] Computer-implemented method for classifying road junctions using GPS data, especially from multiple vehicles in a fleet, comprising the following steps: Using existing GPS data points (10, 11) of vehicle movements, which were recorded in particular by means of GPS sensors of the vehicles at at least one road intersection, Converting the recorded GPS data points into a 2D contour by determining a center point (13) of the road intersection and creating a polygon (12) which represents a 2D contour of the road intersection, Encoding the 2D contour of the road intersection using Elliptic Fourier Descriptors (EFDs) via a Fourier transformation procedure and classifying the encoded data into different intersection types. [2] Method according to claim 1, wherein the classification into different crossing types is carried out by a trained Deep Neural Network (DNN). [3] Method according to claim 2, wherein the Deep Neural Network (DNN) is trained on the intersection types of a T-junction, an X-junction and / or a roundabout. [4] Method according to one of the preceding claims, wherein the 2D contour of the road intersection is formed by an iterative calculation of vertices of the polygon (12) around the center point (13) of the intersection. [5] Method according to one of the preceding claims, wherein the Fourier coefficients are calculated by a forward transformation of the 2D contour (12) of the road intersection. [6] Method according to any of the preceding claims, wherein the Fourier coefficients are encoded into a fixed matrix of numbers to ensure a uniform input size for the DNN. [7] Method according to any of the preceding claims, wherein the DNN classifies road intersections in real time or near real time. [8] Method according to one of the preceding claims, wherein the road junctions are entered in a digital map with the classified junction types. [9] Method according to one of the preceding claims, wherein the collected GPS data (10, 11) are supplemented by sensor data from other sensors, such as a video camera, a radar sensor and / or a lidar sensor. [10] Method according to any of the preceding claims, wherein the classified intersection types are used to improve a navigation system to assist a driver and / or to improve a control system of a self-driving or semi-autonomous vehicle. [11] Method according to any of the preceding claims, wherein the encoded data are used to continuously improve the DNN by using the new encoded data as training data for the DNN and / or by automatically detecting new crossing types and training and adding new DNNs for these crossing types using the associated training data of these new crossing types. [12] Method according to one of the preceding claims, wherein the GPS data points (10, 11) are filtered based on predefined criteria, such as a speed of the vehicle, a direction of travel and / or a distance of the vehicle to the center point (13) of the road intersection, so that faulty GPS data points are filtered out to generate precise 2D contours. [13] Method according to one of the preceding claims, wherein an iterative calculation of the vertices of the polygon (12) is carried out by a weighted consideration of the GPS data points (10, 11) depending on the movement of the vehicle and / or the direction of travel of the vehicle. [14] Method according to any of the preceding claims, wherein the Fourier transform is optimized such that at most the first ten elliptic Fourier descriptors (EFDs) are used for encoding, minimizing the amount of data and ensuring minimal distortion of the contour compared to an actual contour of the road intersection. [15] Device for carrying out the method according to any one of claims 1 to 14, comprising: a processing unit (20) which is set up to process GPS data points (10, 11) of vehicle movements at at least one road intersection and to convert the data into a 2D contour by determining a center point (13) and creating a polygon (12), a storage unit (21) configured to store the calculated 2D contours and / or coded Fourier coefficients, a classification unit (22) which is set up to classify the encoded data of the road intersection contour into different intersection types using a trained deep neural network (DNN), and a communication interface (23) which is set up to transmit the classified intersection types to a navigation system and / or a control system of a vehicle.

Images