Method and system for distance determination

Abstract

Method for distance determination, in which at least one image of a scene is determined by means of at least one camera, and in which at least one feature (53) of the at least one image is recognized by means of a pattern recognition method and at least one mathematical function with which the at least one feature (53) of the at least one image can be modeled is determined and in which the mathematical function (17) is modulated onto at least one predetermined reference image (30),and in which the pattern recognition method is applied to the at least one reference image (31) modulated by the mathematical function (17) and a geometric deviation between a position of at least one feature (41) detected by the pattern recognition method in the modulated reference image and a position of at least one feature (45) corresponding to the at least one detected feature in the at least one unmodulated reference image (30) is minimized by adjusting function parameters of the pattern recognition method, and in which the pattern recognition method is repeated on the basis of the image determined by the at least one camera using the adjusted function parameters, and in which a distance of at least one feature (55) detected during the repeated pattern recognition method to the at least one camera is determined.

Description

The present invention relates to a method and a system for distance determination. Methods based on triangulation principles are used to determine the distance between a camera and an object. Features such as the object's edges are extracted from an image captured by the camera and used as reference points to determine the distance between the camera and the object using a triangulation process. Since optical effects, such as glare effects, can cause difficulties in extracting specific features from an image when using a camera, it may happen that determining the distance between the camera and an object is not possible or not optimal. The known prior art documents are US 6 658 149 B1 , US 2003 / 0 053 660 A1 , US 2012 / 0 050 074 A1 , US 2015 / 0 145 959 A1 and WO 2012 / 145 819 A1 . Against this background, it is an object of the present invention to provide a method that enables the determination of a distance from a camera to an object even under difficult optical conditions. To solve the aforementioned problem, a method for distance determination is presented in which at least one image of a scene is determined using at least one camera, and in which at least one feature of the at least one image is recognized using a pattern recognition method, and at least one mathematical function is determined with which the at least one feature of the at least one image can be modeled, and in which the mathematical function is modulated onto at least one given reference image.and in which the pattern recognition method is applied to the at least one reference image modulated by the mathematical function, and a geometric deviation between a position of at least one feature detected by the pattern recognition method in the at least one modulated reference image and a position of at least one feature corresponding to the at least one detected feature in the at least one unmodulated reference image is minimized by adjusting function parameters of the pattern recognition method, and in which the pattern recognition method is repeated for the image determined by the at least one camera using the adjusted function parameters, and a distance of at least one feature detected during the repeated pattern recognition method to the at least one camera is determined. Embodiments of the presented invention are described in the description and the dependent claims. The presented method is particularly useful for determining the distance of a feature of an object in a scene to an observer, such as a camera. According to the invention, a triangulation method performed using a camera image is validated or modified by validating or modifying a pattern recognition method for recognizing the features underlying the triangulation method. To validate or modify a pattern recognition process implemented in a vehicle, for example, it is intended that the respective functional parameters, i.e., settings such as discrimination or threshold values ​​of the pattern recognition process, are checked against a reference image and optimized if necessary. Accordingly, it is intended that features such as edges or individual pixels are first detected in an image acquired by a camera using a pattern recognition process executed with standard functional parameters. The features detected by the pattern recognition process are then modeled, i.e., described using a mathematical function, such as a 2D interpolation function. The mathematical function, which can be used to model features detected in a real image captured by a camera using the pattern recognition method provided in the invention, is modulated, according to the invention, onto a reference image, i.e., a predetermined artificially generated image whose optical features are known. By modulating the reference image, the respective optical effects detected in an image acquired by the camera, such as glare effects or similar interference effects described by the mathematical function, can be mathematically mapped onto the reference image, i.e., modulated onto the reference image. This means that by modulating the mathematical function onto the reference image, the respective effects detected in the image acquired by the camera are transferred to the reference image.By modulating the mathematical function onto the reference image, the reference image changes. For example, modulation using a Gaussian normal distribution of brightness values ​​as a mathematical function transforms sharp contrasts between white and black areas in the reference image into gradually blended shades. Once the reference image has been modulated using the mathematical function, the pattern recognition procedure can be applied to the modulated reference image using the standard function parameters, i.e., the respective function parameters used by a vehicle for the original execution of the pattern recognition procedure. Since the original optical features of the reference image are known, a deviation between features detected in the modulated reference image using the pattern recognition procedure and the respective known optical features of the unmodulated reference image can be calculated. The known features in the reference image correspond to the features detected in the modulated reference image using the pattern recognition procedure. Based on the deviation between the known optical features of the reference image and the features detected on the modulated reference image, the respective functional parameters of a feature recognition method used to detect the features can be validated by classifying the functional parameters as OK, for example, if deviations between the known optical features and the features detected on the modulated reference image in, for example, a position, a diameter, an energy density, or any other technically suitable measure for describing the optical features are below a predefined threshold or within a predefined variance. To modify the functional parameters of a particular feature recognition method, the parameters can be changed, and a new feature recognition process can be performed based on these modified parameters. The parameters can be changed until the respective deviations between the features of the modulated reference image recognized using the adjusted parameters and the known features of the reference image corresponding to the recognized features are below a predefined threshold or within a predefined variance. Once adapted function parameters have been determined, i.e., function parameters that lead to deviations between respective features known in an unmodulated reference image and features detected in a modulated reference image, which are below a predetermined threshold or within a predetermined variance, the adapted function parameters can be used to determine or measure a distance between the detected features and an observer, e.g., using a triangulation method. In one possible embodiment of the presented method, the mathematical function represents an edge profile as the at least one feature of the at least one image, and an edge profile corresponding to the edge profile of the at least one reference image is modulated by means of the mathematical function. If a pattern recognition method used in a vehicle is not optimally configured with respect to its functional parameters, the positions of edges or features in a camera-captured image, as determined by the pattern recognition method, will not correspond to the edges or features in a reference image. To achieve at least an approximate alignment of the edge positions or features determined by the pattern recognition method with those in the reference image, the functional parameters used by the pattern recognition method to detect the position of the edges or features in the image are adjusted according to the invention. It is particularly intended that the mathematical function provided according to the invention is selected such that it maps an edge profile, i.e., in particular, the profile of the perimeter of a feature. By applying such a mathematical function to a corresponding feature of a respective reference image, the feature of the reference image is modified according to the edge profile of the feature of the image, i.e., the real image captured by the camera. The mathematical processes used to modify the feature of the reference image are known. Accordingly, an artificial image can be generated by means of such a mathematical function, which corresponds to an image captured by a camera, but for which the exact edge profiles of the respective features are known. In one possible embodiment of the presented method, the respective function parameters used by the pattern recognition method to detect the at least one feature, together with the at least one mathematical function for modeling the feature, are transmitted via a communication interface from a vehicle containing the camera to a server. The server is designed to determine the adapted function parameters and transmit them to the vehicle via a communication interface. To execute the presented procedure as quickly as possible and to conserve the computing resources of a vehicle configured to perform the procedure, the presented procedure is designed so that computationally intensive operations are performed on a server, i.e., an external computing unit. For this purpose, the data required to perform these computationally intensive operations is transmitted to the server via a communication interface provided by the vehicle. For example, data required for modulating the reference image using the mathematical function and / or for adjusting the function parameters, such as standard function parameters used by the vehicle and / or the mathematical function itself, can be transmitted from the vehicle to the server.A cable or any technically suitable wireless interface, such as a Bluetooth interface or a WLAN interface, can serve as the communication interface. In another possible embodiment of the presented method, it is provided that the reference image is an image whose optical properties with respect to the feature modeled by means of the at least one mathematical function are known. To enable the most optimal adaptation of function parameters, it is necessary to detect deviations between at least one feature detected in the modulated reference image using the pattern recognition method and at least one known feature corresponding to the detected feature in the reference image. This means that the respective features detected using the pattern recognition method, such as the position or energy density of the given pattern, must be known so that the influence of the mathematical function on the given image or reference image can be precisely described mathematically, and the properties of each feature are also known in the reference image modulated by the mathematical function. By comparing a feature detected on a modulated reference image using the pattern recognition method with a known feature in the reference image corresponding to the detected feature, for example,A difference between the features is determined, which corresponds to an inaccuracy of the pattern recognition method due to optical interference. In another possible embodiment of the presented method, it is provided that at least one optical feature from the following list of optical features or a combination thereof is selected as the feature: edges, single pixels, circles, shading. In another possible embodiment of the presented method, it is provided that the mathematical function by which the at least one feature is modeled is chosen depending on at least one optical property of the at least one feature. The mathematical function is used in particular to model so-called "light characteristics," i.e., characteristics that describe a light distribution. Accordingly, the mathematical function is chosen such that a given light characteristic, such as an energy density at a specific point, is described with particular mathematical precision. For example, a normal distribution, i.e., a Gaussian normal distribution, is suitable for describing an energy density at a specific point in an image. In a further possible embodiment of the presented method, it is provided that, for selected images from a large number of images acquired by means of the at least one camera, adapted functional parameters are determined. It is further provided that the at least one spotlight is adjusted using the currently adapted functional parameters. To dynamically adapt a headlight to the current environment and enable precise adjustment using function parameters adapted to the respective conditions, it is conceivable that the presented method is repeated and the function parameters are continuously modified or adjusted. For this purpose, for example, individual images containing particularly striking features—i.e., features whose optical properties exceed a predefined threshold—can be used to carry out the presented method. In another possible embodiment of the presented method, it is provided that for selected images of a large number of at least one spotlight, adapted functional parameters are determined using an average value of the respective images obtained, and the adapted functional parameters are set. To avoid continuous high-frequency readjustment of a given headlight, it may be provided that the headlight is adjusted using respective average values ​​of, for example, adapted respective functional parameters determined in a specified time range. In another possible embodiment of the presented method, it is provided that the server determines adapted function parameters for several vehicles grouped in a geographical area, whereby optical conditions in the geographical area are inferred from deviations between the function parameters transmitted to the server by the respective vehicles of the grouped vehicles and the adapted function parameters determined by the server for the respective vehicles, and wherein the optical conditions are taken into account in a future determination of adapted function parameters for vehicles in the geographical area. By modifying a multitude of function parameters, a pattern analysis can be used, for example, to identify a disturbance factor in a given geographical area. Once such a disturbance factor, such as a low sun, is known, subsequent calculation methods for adjusting the headlights of various vehicles can be optimized taking the disturbance factor into account, for example by modifying a reference image with a mathematically modeled disturbance factor. Furthermore, the present invention relates to a distance determination system comprising a server and a control unit, wherein the control unit is configured to recognize at least one feature in at least one image of a scene captured by at least one camera using a pattern recognition method and to determine a mathematical function with which the at least one feature of the at least one image can be modeled, and wherein the control unit is further configured to transmit the at least one mathematical function together with function parameters used by the pattern recognition method to the server via a communication interface, and wherein the server is configured to modulate at least one reference image using the at least one mathematical function, and wherein the server is further configured toto apply the pattern recognition method with the transmitted function parameters to the at least one reference image and to minimize a deviation between a position of at least one feature recognized by the pattern recognition method in the at least one modulated reference image and a position of at least one feature corresponding to the at least one recognized feature in the at least one unmodulated reference image by adjusting function parameters of the pattern recognition method and to transmit correspondingly adjusted function parameters to the control unit, and wherein the control unit is further configured toto repeat the pattern recognition procedure for the at least one image acquired by means of the at least one camera using the adapted function parameters and to determine a distance of at least one feature recognized during the repeatedly performed pattern recognition procedure to the at least one camera. The presented system serves in particular to carry out the presented procedure. Specifically, the presented system may include a vehicle whose headlights are adjusted or calibrated using the presented procedure. Further advantages and embodiments of the invention will become apparent from the description and the accompanying drawings. It is understood that the features mentioned above and those to be explained below can be used not only in the combinations specified, but also in other combinations or individually, without departing from the scope of the present invention. Fig. 1 shows a schematic representation of a possible embodiment of the method according to the invention. Fig. 2 shows a mathematical function as provided according to a possible embodiment of the method according to the invention. Fig. 3 shows a representation of a predetermined image modified by means of a mathematical function, as provided according to a possible embodiment of the method according to the invention. Fig. 4 shows a process for optimizing function parameters according to a possible embodiment of the method according to the invention.Figure 5 shows a comparison of a feature identified according to the prior art and a feature identified according to a possible embodiment of the method according to the invention. Figure 1 shows a schematic sequence of the method according to the invention by means of a flowchart 1. In a recording step 3, an image of a scene or the vehicle's environment, into which a pattern has been projected, is recorded using a camera, which is, for example, part of a vehicle. Due to a surface structure of the scene's environment, deformations of the pattern in the image result. Based on the image obtained in the recording step 3, at least one feature, in particular a feature that describes a light distribution or light incidence in the image, such as an energy density, is modeled or mathematically described in a modeling step 5 using a mathematical function, such as a 2D interpolation function. A given image, i.e. a reference image, which is in particular black and white and whose optical properties are known, is convolved or changed in an adjustment step 7 with the mathematical function, i.e. modulated by means of the mathematical function, so that an influence of the feature on the reference image is artificially or mathematically generated or simulated. The reference image itself, unlike the real image captured by the camera, is a synthetic, i.e., artificially produced image. A pattern recognition procedure is then performed on the modulated reference image based on standardized or predefined function parameters, so that a pattern or corresponding features are recognized on the modulated reference image. Since the optical properties of the unmodulated reference image are known, the patterns or features of the modulated reference image recognized by the pattern recognition procedure can be assigned to corresponding patterns or features in the unmodulated reference image. Based on the corresponding patterns or features from the modulated and the unmodulated reference images, a deviation between the patterns or features recognized by the pattern recognition procedure from the modulated reference image and the corresponding patterns or features from the unmodulated reference image can be determined in a determination step 9. In adaptation step 11, the function parameters are determined that lead to a pattern recognition procedure which identifies the pattern or the corresponding features as precisely as possible, i.e., where the deviation between the pattern or features recognized by the pattern recognition procedure and a pattern or respective features of the unmodulated reference image is minimal. For this purpose, the pattern recognition procedure can be repeatedly executed with various modified or adapted function parameters and / or a predefined threshold value can be used, below which the deviation is deemed suitable. The adjusted function parameters are used in setting step 13 to perform a pattern recognition procedure in a given vehicle. This allows for an optimized and highly accurate triangulation procedure, based on these parameters, using points projected into the vehicle's surroundings, for example, using a headlight. The triangulation procedure then determines the distance between each feature detected in the vehicle's environment and the camera. This distance can then be used, for example, to adjust or calibrate the headlight. Figure 2 shows a scene 15 recorded by a camera, which is modulated by a mathematical function 17, as illustrated by a visualization of the mathematical function 17 in diagram 19. Diagram 19 shows a visualization of the energy density distribution in scene 15, which corresponds to a Gaussian normal distribution. The energy density is particularly high in a center 21 of scene 15, as illustrated by a peak 23. Figure 3 shows a predefined image, i.e., a reference image 30, which is folded or modified by the mathematical function 17 from Figure 2 to form a modulated reference image 31, as indicated by arrow 33. The mathematical function 17 transforms edges 35 of the reference image 30 into blurred areas 37 in the modified image. Figure 4 illustrates a modification procedure for the function parameters of a pattern recognition method. On the modulated reference image 31, as described in Figure 3, a pattern recognition procedure is performed in the left region of Figure 4. This procedure detects patterns in the form of lines 41 at the positions indicated by arrows 43. Since the correct position of the patterns in the reference image 30 is also known in the modulated reference image 31 due to the known mathematical operation to be performed, namely the previously described convolution, as indicated by lines 45 and arrows 47, any deviation caused by inaccuracies in the pattern recognition procedure between the detected patterns according to lines 41 and the correct patterns according to lines 45 can be determined. Once the deviation is known, new, optimized function parameters for the pattern recognition procedure are determined. The pattern recognition process is repeated, as shown in the right-hand section of Fig. 4. With the new function parameters, the corresponding patterns according to lines 49 are now recognized, whose positions correspond to the known patterns according to lines 45 or exhibit a minimal distance to lines 45, as indicated by arrows 51. Accordingly, the pattern recognition process can be adaptively adjusted to a specific scene or to the specific lighting conditions affecting a scene. The optimized function parameters can be selected based on the determined distance or randomly and verified by repeating the pattern recognition process. If any adjusted function parameters result in a greater distance than previously known parameters, they are discarded. The pattern recognition process can be repeated until a minimum distance or a distance below a predefined threshold is determined. Figure 5 shows a comparison of a feature 53, detected using a pattern recognition method based on standardized function parameters, and a feature 55, detected using a pattern recognition method based on function parameters optimized according to a possible embodiment of the presented method. It is evident that feature 55 is detected significantly more precisely than feature 53. Consequently, significantly more accurate triangulation for distance determination can be performed using feature 55 than is possible using feature 53.

Claims

Method for distance determination, in which at least one image of a scene is determined by means of at least one camera, and in which at least one feature (53) of the at least one image is recognized by means of a pattern recognition method and at least one mathematical function with which the at least one feature (53) of the at least one image can be modeled is determined and in which the mathematical function (17) is modulated onto at least one predetermined reference image (30),and in which the pattern recognition method is applied to the at least one reference image (31) modulated by the mathematical function (17) and a geometric deviation between a position of at least one feature (41) detected by the pattern recognition method in the modulated reference image and a position of at least one feature (45) corresponding to the at least one detected feature in the at least one unmodulated reference image (30) is minimized by adjusting function parameters of the pattern recognition method, and in which the pattern recognition method is repeated on the basis of the image determined by the at least one camera using the adjusted function parameters, and in which a distance of at least one feature (55) detected during the repeated pattern recognition method to the at least one camera is determined. Method according to claim 1, wherein the mathematical function (17) maps an edge profile as the at least one feature (53) of the at least one image and an edge profile corresponding to the edge profile of the at least one reference image (31) is modulated by means of the mathematical function (17). Method according to claim 1 or 2, wherein at least one optical feature from the following list of optical features or a combination thereof is selected as the feature to be modeled: edges, single pixels, circles, shading. Method according to one of the preceding claims, wherein the mathematical function (17) by means of which the at least one feature is modeled is selected depending on at least one optical property of the at least one feature. Method according to one of the preceding claims, wherein a normal distribution is chosen as the mathematical function (17) by means of which the at least one feature is modeled. A method according to one of the preceding claims, wherein for selected images of a plurality of images obtained by means of the at least one camera, adapted functional parameters are determined, and wherein at least one headlight of a vehicle comprising the at least one camera is adjusted using currently adapted functional parameters. Method according to claim 6, wherein for selected images of a plurality of images obtained by means of the at least one camera, appropriately adapted functional parameters are determined, and wherein the at least one headlight is adjusted using an average value of the appropriately adapted functional parameters. Method according to claim 6 or 7, wherein each of the function parameters used by the pattern recognition method to recognize the at least one feature, together with the at least one mathematical function (17) for modeling the feature, are transmitted from the vehicle to a server via a communication interface, and wherein the server determines the adapted function parameters and transmits them to the vehicle via a communication interface. The method of claim 8, wherein the server determines adapted function parameters for several vehicles grouped in a geographical area, and wherein, based on deviations between function parameters transmitted to the server by the respective vehicles of the grouped vehicles and adapted function parameters determined by the server for the respective vehicles, conclusions are drawn about optical conditions in the geographical area, and wherein the optical conditions are taken into account in a future determination of adapted function parameters for vehicles in the geographical area. Distance determination system comprising a server and a control unit, wherein the control unit is configured to recognize at least one feature (53) in at least one image of a scene illuminated by at least one spotlight, captured by at least one camera, using a pattern recognition method, and to determine a mathematical function (17) with which the at least one feature (53) of the at least one image can be modeled, and wherein the control unit is further configured to transmit the at least one mathematical function (17) together with function parameters used by the pattern recognition method to the server via a communication interface, and wherein the server is configured to modulate at least one reference image (30) using the at least one mathematical function (17), and wherein the server is further configured toto apply the pattern recognition method with the function parameters transmitted by the vehicle on the at least one reference image (30) and to minimize a deviation between a position of at least one feature detected by the pattern recognition method in the at least one modulated reference image (31) and a position of at least one feature corresponding to the at least one detected feature in the at least one unmodulated reference image (30) by adjusting function parameters of the pattern recognition method and transmitting correspondingly adjusted function parameters to the vehicle, and wherein the control unit is further configured toto repeat the pattern recognition procedure using the at least one image obtained by means of the at least one camera and the adapted function parameters, and to determine a distance of at least one feature (55) detected during the repeated pattern recognition procedure to the at least one camera.

Images