Method for determining contour of object

Abstract

The present application provides a method for determining a contour of an object with respect to a sensor configured for providing sensor detection results, each sensor detection result comprising a corresponding position at the object. An initial profile is determined that includes a plurality of segments, each segment relating to a corresponding initial subset of the sensor detection results. A corresponding one of the sensor detection results is associated with each section of the initial profile. A corresponding weight is set for each sensor detection result, each weight depending on a relative position of the sensor detection result with respect to the associated section. Each section of the initial contour is refined by using the weights of the sensor detection results in the corresponding surrounding group associated with the corresponding section in order to determine a final contour of the object comprising the refined sections.

Description

Technical Field

[0001] The present disclosure relates to a method for determining a contour of an object with respect to a sensor. Background Art

[0002] Different methods are known for localizing the contour or boundary of an object based on a point cloud provided by a ranging sensor such as a radar sensor or a Lidar sensor. These methods often rely on two-dimensional point clusters or point clouds given in a bird's-eye view representation. Known methods include, for example, methods relying on convex hull, concave hull, recasting alpha shape, or Delaunay triangulation.

[0003] Known methods for determining the contour or boundary of an object based on a point cloud or point cluster provided by a ranging sensor are often computationally expensive, and they generally do not take into account the statistical distribution or spread of the point cloud. Such spread may be caused by noise and / or clutter of the ranging sensor (e.g., a radar sensor or a Lidar sensor).

[0004] Therefore, the boundary of a real object, especially the edge of a real object, is often represented by a blurred contour because known methods generally localize such a contour in a conservative manner by generally connecting elements or detections in the point cloud having the closest distance with respect to the sensor. As a result, outlier points may be overestimated. In summary, the contour of an object determined by known methods cannot accurately represent the boundary of a real object in many cases.

[0005] Therefore, there is a need for a method capable of accurately localizing the boundary of an object based on the detection results of a ranging sensor. Summary of the Invention

[0006] The present disclosure provides a computer-implemented method, a computer system, and a non-transitory computer-readable medium according to the independent claims. Embodiments are given in the dependent claims, the description, and the drawings.

[0007] In one aspect, the present disclosure relates to a computer-implemented method for determining a contour of an object with respect to a sensor, wherein the sensor is configured to provide a plurality of sensor detection results, each sensor detection result including a corresponding position at the object. According to the method, an initial contour of the object including a plurality of segments is determined, wherein each segment is related to a corresponding initial subset of the sensor detection results. A respective surrounding set of the sensor detections is associated with each segment of the initial contour. A corresponding weight is set for each of the sensor detections in the respective surrounding set, wherein each weight depends on the relative position of the sensor detection result with respect to the associated segment. Each segment of the initial contour is refined by using the weights of the sensor detections in the respective surrounding set associated with the corresponding segment in order to determine a final contour of the object including the refined segments.

[0008] The sensor can be a ranging sensor such as a radar sensor or a lidar sensor, which is capable of determining a point at the object or the range or distance of a detection result with respect to a reference position at the sensor. In addition, such a sensor is also capable of estimating an angle, for example, with respect to a reference line passing through the sensor along the longitudinal axis of a vehicle on which the sensor is mounted, for example. Thus, the sensor is capable of determining the spatial position of the corresponding point or detection result at the object in two or three dimensions. For example, if a radar sensor is used, the sensor detection results can be represented in a two-dimensional coordinate system in a bird's-eye view, that is, viewing the sensor and the object under consideration from above.

[0009] Therefore, determining the contour of the object with respect to the sensor (i.e., the initial contour and the final contour) means that the contour or boundary of the object can be represented in a bird's-eye view in such a two-dimensional coordinate system. The center of such a coordinate system can be located at the sensor.

[0010] Determining the corresponding range of the sensor detection results and their respective angles with respect to the reference position and reference direction at the sensor can be performed by a processing unit that can be installed in a vehicle on which the sensor is mounted, for example. In addition, such a processing unit can also be configured to perform the steps of the above method. Thus, the sensor detection results can be obtained with respect to the external environment of such a vehicle, and each sensor detection result can be associated with an object located in the external environment of the vehicle.

[0011] According to this method, an initial subset in the sensor detection results is selected and considered to determine an initial contour, rather than generally connecting the nearest points or detection results when observing from the sensor (which is mainly the basis of known methods). Since a subset is selected to determine the initial contour, a certain spread or distribution of the detection results is automatically incorporated into the method. Compared with the results of known methods, this can allow for determining a contour closer to the true contour of the object.

[0012] In addition, since the initial subset in the sensor detection results is selected by, for example, pre-defined criteria and the corresponding surrounding groups in the sensor detection results are selected for the refinement of the segments, outliers in the sensor detection results can be identified and excluded from the process of estimating the contour of the object. Therefore, the accuracy of the final contour can be improved by avoiding the influence of obvious outliers.

[0013] Compared with the results of known methods, the initial contour line can have improved smoothness because a certain distribution of the sensor detection results is represented by the corresponding initial subset on which the corresponding segments for determining the initial contour are based. However, if the sensor detection results have a large lateral spread (e.g., relative to the initial contour), then this initial contour may not correctly represent the true contour or boundary of the object. For example, if when observing from the sensor, the sensor detection results have a large spread in both the longitudinal and lateral directions, the initial contour may still extend, e.g., through the middle of the cloud or points of the points or sensor detection results.

[0014] Compared with the true contour of the object, such an initial contour would have a greater distance from the sensor if all the sensor detection results belonging to the corresponding subset are considered in the same or equivalent way. In other words, the sensor detection results belonging to the corresponding initial subset may all have the same weight for the corresponding segments for determining the initial contour.

[0015] Therefore, in some cases where there is a large spread of the sensor detection results and even some skewness or inclination of the point cloud of the sensor detection results (which may cause the initial contour to deviate from the true contour of the object), the initial contour needs to be refined. Therefore, the corresponding surrounding groups of the sensor detection results are associated with each segment of the initial contour, where the association of the sensor detection results of the corresponding surrounding groups can be performed by different criteria described below.

[0016] Within the sensor detection results corresponding to the surrounding groups, each sensor detection result can be set to have a corresponding weight, which depends on the relative position or distance of the sensor detection result with respect to the associated segment belonging to the initial contour. For example, a sensor detection result in the corresponding surrounding group that has a relatively far distance from the initial contour and / or can be positioned closer to the sensor than the initial contour can be set to have a weight that is greater than the weight of a sensor detection result that is closer to the initial contour and / or can have a relatively farther distance from the sensor with respect to the initial contour.

[0017] Each segment of the initial contour can be refined based on the weights of the detection results within the corresponding surrounding group associated with the corresponding segment. For example, by translating the segment based on the weights, or by performing a certain regression process based on the sensor detection results belonging to the corresponding surrounding group and having different weights due to different relative positions with respect to the associated segment. Compared with the method steps for determining the initial contour and its segments, the refinement of each segment of the initial contour based on the weights of the detection results can incorporate the spread or distribution of the positions of the sensor detection results with respect to the sensor in a more realistic manner. Therefore, the final contour including the refinement can represent the true contour of the object more accurately than the initial contour.

[0018] According to an embodiment, the initial contour may further include a plurality of vertices, and the segments of the initial contour may extend between a corresponding pair of the vertices. One of the sensor detection results can be selected as the first vertex, and the other vertices after the first vertex can be iteratively determined by selecting a corresponding initial subset of the sensor detection results with respect to the corresponding previous vertex, and by estimating the position of the next vertex using the corresponding initial subset selected with respect to the previous vertex.

[0019] In other words, the first vertex can be selected from the sensor detection results (e.g., a sensor detection result on one of the outer sides along the lateral direction of the point cloud representing the sensor detection results when observed from the sensor), and the second vertex can be determined based on the initial subset selected for the first vertex. For example, within the environment near the first vertex, by using the subset of the sensor detection results to determine the direction and size of the first segment extending from the first vertex to the second vertex. Thereafter, the initial subset associated with the second vertex can be determined by selecting the sensor detection results having positions close to the second vertex, and the initial subset associated with the second vertex can be used to determine the third vertex, and so on. In this way, a polyline including the segments of the initial contour can be iteratively drawn in the point cloud representing the sensor detection results.

[0020] By selecting a corresponding initial subset relative to a corresponding previous vertex, some extensions of the detection results can be considered when determining the next vertex. However, outliers that may be located far from the corresponding vertex and may interfere with the correct determination of the polyline representing the segment for the initial contour can be excluded. This can improve the accuracy of the initial contour, i.e., the accuracy of the polyline including the segments connected to each other at the corresponding vertex.

[0021] Selecting a corresponding initial subset in the sensor detection results for a previous vertex can include selecting the sensor detection results located within a predefined region around the previous vertex. For example, the corresponding previous vertex can be the center of the predefined region. Thus, the predefined region can be iteratively shifted from one vertex to the next for selecting the corresponding initial subset in the sensor detection results in order to determine additional vertices again. This method step of iteratively selecting the sensor detection results and determining the corresponding next vertex can require low computational effort.

[0022] Therefore, the predefined region can be a rectangular bounding box, or even a square bounding box, which can be centered at the corresponding vertex and can be shifted from the corresponding one of the vertices to subsequent vertices when iteratively determining multiple vertices. The rectangular or square bounding box can be represented in a bird's-eye view relative to the sensor and the object. For example, when using a radar sensor, the radar sensor can provide sensor detection results in two dimensions, i.e., in a plane extending parallel to the ground.

[0023] In addition, the sensor detection results selected for the initial subset selection of the previous vertex can be excluded from the initial subset of the subsequent vertex. Thus, the sensor detection results that have been used to locate or determine the vertices of the initial contour may not be used again to determine additional vertices. Thereby, the accuracy of the initial contour can be improved because, without reusing the sensor detection results, the initial contour can more correctly follow the trajectory of the sensor detection results.

[0024] In addition, the sensor detection results selected as the first vertex can also be confirmed in the following way: determining at least two sorted lists for the positions of the sensor detection results, each sorted list referring to the corresponding coordinates of the positions, selecting the sorted list with the largest difference between the first element and the last element in the list, and selecting the first element in the selected sorted list as the first vertex.

[0025] The location of the sensor detection result may include coordinates, which may be represented relative to a coordinate system centered at or near the reference position of the sensor, and if the coordinate system is defined in a two-dimensional manner (e.g., in a bird's-eye view), the corresponding coordinates of the location of the sensor detection result may include an x coordinate and a y coordinate relative to such a coordinate system. The maximum extent of the sensor detection result along one of the axes of the coordinate system may determine the direction or alignment of the cluster or point cloud of points representing the sensor detection result. Due to these sorting and selection steps, a suitable sensor detection result near the outer edge of such a point cloud representing the sensor detection result may be selected as the first vertex or starting point of the polyline representing the initial contour.

[0026] Estimating the position of the next vertex by utilizing the corresponding initial subset may include determining the segment vector extending from the previous vertex to the next vertex by: calculating the geometric mean of the sensor detection results in the corresponding initial subset, where the geometric mean may provide the direction of the segment vector, and determining the sensor detection result furthest from the previous vertex within the corresponding initial subset, where the distance between the furthest sensor detection result and the previous vertex may define the absolute value of the segment vector.

[0027] Since the segment vector may connect a pair of vertices, i.e., the previous vertex and the next vertex, the steps of calculating the geometric mean and determining the furthest sensor detection result may be used to iteratively provide the next vertex. Additionally, the geometric mean may reflect the distribution or extent of the sensor detection results within the corresponding initial subset.

[0028] According to a further embodiment, the corresponding main region may be arranged symmetrically with respect to the corresponding segment, and the sensor detection results of the corresponding surrounding group may include regular sensor detection results located within the main region. Additionally, the corresponding modified region may be determined by modifying the corresponding main region according to the distribution of the sensor detection results relative to the sensor, and the sensor detection results of the corresponding surrounding group may further include special sensor detection results located within the modified region and outside the main region. The regular sensor detectors may be set to have a normal weight, while the special sensor detection results may be set to have an increased weight greater than the normal weight.

[0029] Thus, the sensor detection results in the corresponding surrounding group associated with the corresponding segment of the initial contour may include two sets of different sensor detection results, namely, the regular sensor detection results set to have normal weights and the special sensor detection results set to have increased weights relative to the normal weights. Thus, the sensor detection results located outside these two regions (i.e., outside all the main regions and the modified regions) may not be associated with any segment of the initial contour, such that these detection results can be identified as outliers.

[0030] The corresponding main region may have a predefined shape. For example, it may have the shape of a rectangular bounding box that extends from one vertex to the next along the corresponding segment and has a predefined width smaller than the length of the corresponding segment. The predefined width of such a rectangular bounding box may be based on the characteristics of the sensor, e.g., depending on the resolution of the sensor.

[0031] The modification of the main region depends on the distribution of the sensor detection results relative to the sensor, which can respectively reflect the distribution of the sensor detection results relative to the initial contour. For example, if, as observed from the sensor, the distribution of the sensor detection results has a large spread in one or two directions, it can be expected that the initial contour may not truly reflect the true contour of the object under consideration. Thus, the main region can be modified, e.g., translated in a direction perpendicular to the corresponding segment, in order to provide an asymmetric region relative to the corresponding segment, thereby associating further sensor detection results (i.e., special detection results) that may be located closer to the true boundary of the object under consideration.

[0032] Thus, in the refinement step, the special sensor detection results with increased weights can "attract" the corresponding segment of the initial contour. Thus, due to the influence of the special sensor detection results, the refined segment providing the final contour can follow the true boundary or contour of the object more accurately than the initial contour.

[0033] The corresponding modified region can be determined by translating the corresponding main region in a direction perpendicular to the corresponding segment, and the direction and amount of translation can depend on the distribution of the sensor detection results relative to the sensor. For example, depending on the spread of the sensor detection results, the corresponding bounding box representing the corresponding main region can be translated in the direction towards the sensor or away from the sensor.

[0034] In addition, the corresponding segment vector can be determined as the difference between the position vectors of a pair of vertices associated with the corresponding segment. A corresponding normal vector perpendicular to the corresponding segment vector can be determined for each segment, and the auxiliary vector can be determined as the difference between two other position vectors (i.e., the position vector of the reference near the sensor position and the position vector of the center of the segment closest to the sensor). For all segments, the amount of translation of the main region can be given by the dot product of the auxiliary vector and the normal vector of the segment closest to the sensor, and the direction of translation can be given by the corresponding normal vector of the segment.

[0035] The amount of translation of the main region can be limited by a predefined value, and depending on the dot product of the auxiliary vector and the normal vector of the closest segment, translation can be applied or no shift can be applied to the main region. For example, if the dot product is greater than zero, the translation of the main region of the corresponding segment can be along the direction towards the sensor. Otherwise, no translation can be applied.

[0036] Optionally, the amount of translation is not only related to the absolute value of the translation, but also related to a sign that can be positive or negative. For example, a positive dot product can indicate that the main region is to be translated by an amount provided by the dot product along the direction towards the sensor, while a negative dot product can indicate the opposite, i.e., translating the main region away from the sensor. The translation of the main region can generally be perpendicular to the corresponding segment, i.e., along the corresponding normal vector, and all segments are translated by the same amount, which can be predefined or can depend on the value of the dot product.

[0037] Therefore, a well-defined process can be applied to determine the corresponding modified region belonging to each segment. Such a process can require low computational effort.

[0038] Additionally or alternatively, for each sensor detection result in the corresponding surrounding group, it can be detected whether the sensor detection result is at the initial contour, on the sensor-facing inner side of the initial contour, or on the sensor-averted outer side of the initial contour. The sensor detection result at the initial contour is set to have a normal weight, while the sensor detection result on the inner side is set to have an increased weight greater than the normal weight, and the sensor detection result on the outer side is set to have a decreased weight less than the normal weight.

[0039] Thus, the weight of the sensor detection result that is increased or decreased relative to the normal weight can depend on the corresponding position or distance of the detection result relative to the initial contour. In this way, the expansion of the sensor detection result relative to the initial contour can be reflected in the corresponding weight of the sensor detection result. That is to say, if almost all sensor detection results are located close to the initial contour such that there is little expansion of the sensor detection results relative to the initial contour, then all weights can be close to the normal weight, and the normal weight is, for example, 1.0. Thus, in this case, there is little refinement of the initial contour. On the contrary, if the expansion of the sensor detection results relative to the initial contour is large, then many sensor detection results can be set to have weights that are increased or decreased relative to the normal weight (e.g., 1.0). Thus, the initial contour can be strongly refined (intense refinement) in the direction towards the sensor, for example.

[0040] According to a further embodiment, refining each segment of the initial contour can include using a regression process to modify the corresponding segment to the weighted sensor detection results of the corresponding surrounding group of sensor detection results associated with the corresponding segment. Thus, the weighted sensor detection results can be used as nodes of the regression process. Different types of regression processes can be used, such as Deming regression and / or Kalman filtering.

[0041] The regression process can be iteratively applied to the segments of the initial contour starting from the first segment. Thus, the position of the second vertex defining the first segment can be first refined by applying the regression process to the weighted sensor detection results in the first surrounding subset associated with the first segment. Thereafter, the position of the third vertex can be refined by applying the regression process to the weighted sensor detection results in the surrounding subset associated with the second segment, and so on. Iteratively applying such a standard regression process can still require a low computational amount.

[0042] On the other hand, the present disclosure relates to a computer system configured to receive a plurality of sensor detection results from a sensor configured to determine the corresponding position of the sensor detection results and to perform several or all of the steps of the computer-implemented method described herein.

[0043] The computer system can include a processing unit, at least one storage unit, and at least one non-transitory data storage. The non-transitory data storage and / or the memory unit can include a computer program for commanding the computer to perform several or all of the steps or aspects of the computer-implemented method described herein.

[0044] As used herein, terms such as processing units and modules can refer to, be part of, or include the following: application specific integrated circuits (ASICs), electronic circuits, combinatorial logic circuits, field programmable gate arrays (FPGAs), processors (shared, dedicated, or grouped) that execute code, other suitable components that provide the functionality, or combinations of some or all of the above, such as in a system-on-chip. A processing unit can include a memory (shared, dedicated, or grouped) that stores code executed by the processor.

[0045] In another aspect, the present disclosure relates to a vehicle and a computer system as described above, the vehicle including sensors configured to obtain a plurality of sensor detection results and to determine corresponding positions of the sensor detection results.

[0046] In another aspect, the present disclosure relates to a non-transitory computer-readable medium including instructions for performing several or all steps or aspects of the computer-implemented methods described herein. The computer-readable medium can be configured as: an optical medium such as a compact disc (CD) or a digital versatile disc (DVD); a magnetic medium such as a hard disk drive (HDD); a solid state drive (SSD); a read-only memory (ROM) such as a flash memory; or a similar medium. Additionally, the computer-readable medium can be configured as a data memory accessible via a data connection such as an Internet connection. For example, the computer-readable medium can be an online data repository or cloud storage.

[0047] The present disclosure also relates to a computer program for commanding a computer to perform several or all steps or aspects of the computer-implemented methods described herein. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Exemplary embodiments and functions of the present disclosure are described herein in connection with the following drawings, which schematically show:

[0049] Figure 1 Schematically shows the true contour of an object detected by a radar sensor and an approximate contour of the object estimated based on sensor detection results by applying known methods and method steps according to the present disclosure;

[0050] Figure 2 Schematically shows method steps for determining an initial contour of an object based on sensor detection results;

[0051] Figure 2A Schematically shows Figure 2 The first iteration of the iterative process shown;

[0052] Figure 3Schematically shows method steps for associating sensor detection results with corresponding segments of an initial contour;

[0053] Figure 4 Schematically shows method steps for refining segments of an initial contour;

[0054] Figure 5 Schematically shows examples of determining an object's contour with and without association-based refinement;

[0055] Figure 6 Schematically shows a flowchart illustrating a method for determining an object's contour relative to a sensor according to various embodiments;

[0056] Figure 7 Schematically shows a diagram of a contour determination system according to various embodiments, and

[0057] Figure 8 Schematically shows a computer system having a plurality of computer hardware components configured to perform the steps of the computer-implemented methods described herein. Detailed Description

[0058] Figure 1 A schematically depicts a vehicle 100 equipped with a radar sensor 110. The radar sensor 110 has an instrument field of view 112 and is configured to provide sensor detections 114 related to an object 120 in the external environment of the vehicle 100 and the radar sensor 110.

[0059] Each of the sensor detections 114 includes a corresponding spatial position, namely a range measurement relative to a reference position at the sensor 110, and an angle relative to a reference direction provided by the longitudinal axis of the vehicle 100, namely an azimuth angle. More specifically, the radar waves emitted by the radar sensor 110 are reflected at multiple positions near the true contour or boundary 122 of the object 120, and the radar waves reflected at these positions are received by the radar sensor 110, such that a processing unit (not shown) of the vehicle 100 can determine the spatial position, namely the range measurement and azimuth angle of the corresponding radar detection result 114. The radar detection results 114, namely their corresponding spatial positions, are provided in a so-called bird's-eye view relative to the vehicle 100 and the radar sensor 110, namely in a plane extending parallel to the ground on which the vehicle 100 is currently located.

[0060] To, for example, avoid a collision between the vehicle 100 and the object 120, it is necessary to approximate the external boundary or contour 122 based on the radar detection results 114. Figure 1B shows the results of a known method according to the related art. When viewed from the sensor 110 or the vehicle 100, the true contour 122 of the object 120 is approximated by a line connecting the outermost sensor detection results 114. As can be seen from Figure 1 B, the approximate contour 130 according to the related art is represented as a fuzzy boundary line, which shows deviations from the true contour 122 in many regions. Such a conservative approximation 130 can be determined for the true contour 122 of the object 120 in order to avoid any collision of the vehicle 100 with, for example, an edge or a protrusion of the object 120.

[0061] However, many automotive applications may require a more accurate representation or approximation of the true contour 122. Figure 1 C shows the results of the method steps according to the present disclosure, which provide another approximation of the true contour 122 by using the sensor detection results 114 in a different way. The approximation according to the present disclosure is represented by a polyline 140, which extends between vertices 150 iteratively determined based on the sensor detection results 114, which will be described in detail below. The polyline 140 includes a plurality of segments 160, each of which extends between a pair of vertices 150. When comparing Figure 1 C and Figure 1 B, it can be found that, compared with the approximate contour 130 according to the related art, the polyline 140 provided by the method according to the present disclosure approximates the true contour 122 in a more accurate manner.

[0062] Figure 2 The details for iteratively determining the polyline 140 (see Figure 1 ) are schematically illustrated. First, an initial vertex or a first vertex 151 is determined by selecting a suitable one from the sensor detection results 114. To determine or select the first vertex 151, the corresponding length or extent of the point cloud provided by the sensor detection results 114 is determined. For this purpose, the sensor detection results 114 are sorted (sorted) with respect to two axes of a coordinate system, which is defined, for example, in a bird's-eye view and whose origin is located at the sensor position. That is, the x - coordinates and y - coordinates of the sensor detection results 114 are sorted into two corresponding lists, starting with the maximum values of the corresponding coordinates and ending with the minimum values of the corresponding coordinates.

[0063] Thereafter, the extent of each of the two sorted lists is calculated as the difference between the first element and the last element in the corresponding list. The list with the larger extent between the first element and the last element is selected, and the corresponding axis of the coordinate system defines the Figure 2The sorting direction indicated by arrow 210 in []. The first sensor detection result 114 in the selected sorted list (i.e., sorted relative to direction 210) defines the first vertex 151 of the polyline 140 to be determined.

[0064] The first vertex 151 is the starting point of the iterative process 200 for determining the polyline 140 (see Figure 1 ). In the first iteration labeled 201 (see Figure 2 ), a square bounding box 221 is drawn around the first vertex 151. That is, the first vertex 151 represents the center of the square bounding box 221. The square bounding box 221 encloses a pre-defined area in which the first initial subset of the sensor detection results 114 is selected. In other words, the first initial subset includes all the sensor detection results 114 enclosed by the bounding box 221.

[0065] Calculate the geometric mean of all the sensor detection results belonging to the first subset. This is shown in detail in the enlarged portion labeled Figure 2A in the first iteration 201. Calculating the geometric mean of the sensor detection results 114 within the bounding box 221 provides the average point 232. In addition, the farthest sensor detection result 234 is determined, which has the maximum distance relative to the first vertex 151 among all the sensor detection results 114 belonging to the first subset (i.e., within the first bounding box 221).

[0066] The average point 232 and the farthest sensor detection result 234 are used to determine the segment vector 236 extending from the first vertex 151 to the second vertex 152. The average point 232 defines the direction of the segment vector 236 relative to the first vertex 151, while the farthest sensor detection result (detection point) 234 defines the length or absolute value of the segment vector 206. The positions of the first vertex 151 and the second vertex 152 define the first segment 161 of the polyline 140. That is, the first segment 161 extends between the first vertex 151 and the second vertex 152.

[0067] After determining the second vertex 152, the first bounding box 221 is translated to the second vertex 152 such that the second vertex 152 represents the center of the second square bounding box 222 having the same dimensions as the first square bounding box 221. Thereafter, all the sensor detection results 115 that have been used in the first iteration step 201 to determine the second vertex 152 are removed from the second bounding box 222. Thus, in the second iteration step, the remaining sensor detection results 114 enclosed by the second bounding box 222 and not considered in the first iteration step 201 are used.

[0068] In the second iteration 202, the third vertex 153 is determined in the same way as the second vertex 152. That is, in the second iteration 202, a second segment vector 238 is determined, where the direction of the second segment vector 238 is determined by the geometric mean of all sensor detection results 114 enclosed by the second bounding box 222, and thus these sensor detection results 114 belong to the second subset of the sensor detection results 114. For the second subset, that is, for the sensor detection results 114 within the second bounding box 222, the farthest sensor detection result 114 relative to the second vertex 152 is determined again. The length of the second segment vector 238 is given by the distance between the farthest sensor detection result 114 and the second vertex 152. The second segment vector 238 determines the position of the third vertex 153.

[0069] In the third iteration labeled 203, the fourth iteration labeled 204, and the fifth iteration labeled 205, the fourth vertex 154, the fifth vertex 155, and the sixth vertex 156 are determined in the same way as the second vertex 152 and the third vertex 153, respectively. That is, corresponding bounding boxes 223, 224, and 225 are drawn around the corresponding previous vertices 153, 154, and 155. All sensor detection results 114 that have been used in the previous iteration are removed from the corresponding bounding boxes 223, 224, and 225, and the geometric mean and the farthest sensor detection result of the corresponding sensor detection results 114 enclosed by the bounding boxes 223, 224, and 225 are determined respectively, so as to determine the corresponding segment vectors extending to the next vertices 154, 155, and 156.

[0070] The iteration ends when the bounding box translated to the next vertex is empty, that is, all or almost all sensor detection results 114 have been "consumed" to determine the corresponding next vertex. However, there may be sensor detection results 114 that are not enclosed by any of the bounding boxes 121 to 225 in any iteration. These sensor detection results 114 are identified as outliers and will not be further considered when determining the contour of the object 120 (see Figure 1 ).

[0071] Figure 2 The right side of shows the entire polyline 140 determined through iterations 201 to 205, as well as the vertices 151 to 155 and the sensor detection results 114 used to determine the polyline 140.

[0072] However, in some cases, the polyline 140 determined as described in Figure 2 may not yet provide the true contour 122 of the object 120 (see Figure 1) suitable approximation. This is especially the case when the sensor detection result 114 related to the object 120 (i.e., the point cloud of the sensor detection result 114) has a large distribution in two directions or dimensions (e.g., in the direction along the polyline 140 and in the direction perpendicular to the polyline 140).

[0073] In Figure 5 this is illustrated by the scenes marked by A and D. Figure 5 The different scenes of Figure 1 are depicted with respect to the vehicle 100 including the radar sensor 110 (see Figure 5 ). The vehicle 100 is depicted in

[0074] In Figure 5 for the scene marked by D, the sensor detection result 114 has a large extension only in the first direction, but a small extension in the second direction perpendicular to the first direction. Thus, as Figure 2 shown and as described above, the polyline 140 determined by the iterative process 200 extends very close to the trajectory of the sensor detection result 114. On the contrary, the deviation or distance of the corresponding sensor detection result 114 with respect to the polyline 140 is very small. Therefore, in the scene marked by D, the polyline 140 can be regarded as a good approximation of the true contour of the object for which the sensor detection result 114 has been obtained.

[0075] On the contrary, for Figure 5 the scene marked by A in Figure 2 , the point cloud of the sensor detection result 114 has a large extension in two directions (i.e., the longitudinal direction and the lateral direction when viewed from the vehicle 100). The polyline 140 determined by the

[0076] iterative process 200 of Figure 1 extends through the point cloud of the sensor detection result 114 because of the mechanism for iteratively determining the vertex 150 based on the geometric mean of the corresponding subset of the sensor detection result 114 (i.e., the sensor detection result 114 within the corresponding bounding box) as described above. Figure 5 Therefore, there are many sensor detection results 114 on the side of the polyline 14 facing the vehicle 100 and the sensor 110 (see

[0077] ). That is, there are many sensor detection results 114 closer to the vehicle 100 than the polyline 140. These closer sensor detection results 114 are marked by 510 in the scene A of Figure 1), these portions actually protrude from the assumed contour provided by the polyline 140. Thus, some portions of the object 120 represented by the neighboring sensor detection results 510 may pose a risk of collision with the vehicle 100. In other words, compared to what is indicated by the polyline 140 determined by an iterative procedure as shown in Figure 2 , multiple portions of the true contour of the object 120 (see Figure 1 ) may be located at positions closer to the vehicle 100 and the sensor 110. Therefore, if the polyline 140 in the scenario marked A in Figure 5 is assumed to be the true contour of the object, the vehicle 100 may face the risk of at least partially colliding with the object. This is because the large expansion of the sensor detection results 114 is not correctly reflected when determining the polyline 140.

[0078] Therefore, Figure 1 and Figure 2 as well as Figure 5 the polyline 140 in the A and D scenarios in Figure 1 can be regarded as the initial or original contour 140 of the object 120 (see

[0079] Figure 3 and Figure 4 ), which needs to be refined in order to correctly represent or approximate the true contour 122 of the object 120.

[0080] Figure 3 A shows the first association step of the corresponding part of the sensor detection results 114 with the corresponding segment 160 of the initial contour 140 (see Figure 1 and Figure 2 ).

[0081] For the first association step, a corresponding rectangular bounding box 310 is drawn around each segment 160 such that the bounding box extends symmetrically about the segment 160 between a corresponding pair of vertices 150. Thus, the corresponding lengths of these bounding boxes 310 (which may also be labeled as association bounding boxes) correspond to the corresponding distances between a pair of vertices 150. The width perpendicular to the corresponding segment 160 is determined empirically, for example, related to the spatial resolution of the sensor 110.

[0082] Each sensor detection result 114 is located within one of the corresponding rectangular bounding boxes 310, and each sensor detection result 114 is associated with the corresponding section 160 to which the bounding box 310 belongs. The first association step based on the rectangular bounding boxes 310 symmetrically arranged with respect to the section 160 can also be referred to as regular association.

[0083] In addition, a second association step is applied to the sensor detection results 114 to be able to properly account for the large expansion of the sensor detection results 114 in the direction perpendicular to the initial contour 140. As Figure 3 shown in B, the second association step relies on asymmetric rectangular bounding boxes 320, i.e., asymmetric with respect to the corresponding section 160 of the initial contour 140.

[0084] By translating the rectangular bounding boxes 310 used for regular association in the direction towards the vehicle 100 on which the sensors 110 are mounted, rectangular bounding boxes 320 that are asymmetrically arranged with respect to the section 160 of the initial contour 140 are generated. For each associated bounding box 310, the translation direction 330 is determined by first calculating the corresponding section vector that is the difference between the position vectors of subsequent or adjacent vertices 150. Secondly, for each in the section 160, a normal vector is constructed by rotating the corresponding section vector by 90°. The normal vector defines the direction for translating the corresponding bounding box 310. The corresponding direction for translation is indicated by the corresponding arrow 330.

[0085] Thereafter, an auxiliary vector 340 is determined that extends from the center of gravity 105 of the vehicle 100 (i.e., a position close to the location where the sensor 110 is located) to the midpoint of the section 165 having the closest distance relative to the vehicle 100. Thereafter, the dot product between the auxiliary vector 340 and the normal vector of the closest section 165 is calculated, where the normal vector extends parallel to the direction indicated by the arrow 330 belonging to the closest section 165. If the dot product is greater than zero, the corresponding associated bounding box 310 symmetrically arranged with respect to the section 160 (see Figure 3 A) is translated in the direction towards the vehicle 100 indicated by the arrow 330.

[0086] Within the corresponding translated bounding box 320, additional sensor detection results 114 marked with 360 are located, and these sensor detection results 360 are additionally associated with the corresponding section 160 of the initial contour 140 in the second association step. The sensor detection results 350 that have already been associated with one of the sections 160 in the regular association step (see Figure 3 A) are also shown. The association performed based on the translated bounding box 320 (i.e., the second association step) can also be referred to as special association.

[0087] In addition, the sensor detection result 350 associated with the corresponding segment 160 in the first or regular association step is set to have a normal weight, such as 1.0. In contrast, the sensor detection result 360 associated with the corresponding segment 160 in the second or special association step is set to have an increased weight, such as greater than 1.0. Further, the sensor detection result 114 not associated with any segment 150 in the first association step or the second association step can be identified as an outlier.

[0088] According to an alternative embodiment as shown in Figure 3 C, the sensor detection result 114 located at the initial contour 140 is set to have a neutral weight, such as 1.0, while the sensor detection result 114 located inside the initial contour 140 relative to the vehicle 100 or the radar sensor 110 (see Figure 1 ) is set to have an increased weight 370, such as greater than 1.0, and the weight of the sensor detection result 114 located outside and deviated from the vehicle 100 and the radar sensor 110 is set to have a reduced weight 380 relative to the neutral weight, that is, a weight less than 1.0. The absolute values of the corresponding increased weight 370 and reduced weight 38 are illustrated by the corresponding lengths of the arrows 370 and 380 of the symbolized weights.

[0089] Relative to the normal weight (such as 1.0), the amount for increasing or decreasing the weight depends on the distance by which the corresponding sensor detection result 114 is separated from the initial contour 140. However, in order to exclude outliers from the process of determining the contour of an object, an upper limit can be defined for such a distance.

[0090] Figure 4 Illustrated is a step of refining the initial contour 140 based on the weighted sensor detection results 350, 360 associated with the corresponding segment 160 of the initial contour 140. Figure 4 A shows an enlarged portion of the initial contour 140, which includes one segment 160 of the segments 160, and the weighted sensor detection results 350, 360 associated with this segment 160. The associated sensor detection results 350, 360 represent a set of surrounding sensor detection results 114 associated with the corresponding segment 160. Thus, a set of surrounding sensor detection results 114 includes the sensor detection result 350 that is conventionally associated and set to have a normal weight (e.g., 1.0) in the first association step, and the sensor detection result 360 that is associated with the segment 160 and set to have an increased weight greater than 1.0 in the second or special association step (see also Figure 3 B).

[0091] A section 160 of the initial contour 140 is refined by a fitting or regression process for the section that uses a set of surrounding weighted sensor detection results 350 and 360 associated with the corresponding section 160. For regression, Deming regression or a Kalman filter can be used. Generally, the regression process attempts to minimize the sum of the distances 410 between the sensor detection results 114, 350, 360 and the section 160.

[0092] In Figure 4 B, a regression of the section 160 that depends on equal weights of all the sensor detection results 114 is illustrated. Since the weights are equal, the deviation or distance 410 of each of the sensor detection results 114 will remain unchanged during the regression process, which can be recognized by Figure 4 A and Figure 4 B comparison. However, due to two different association steps, namely the conventional association as shown in Figure 3 A and the special association as shown in Figure 3 B, or due to different weights as illustrated in Figure 3 C, different weights are used for the sensor detection results 114 according to the method of the present disclosure.

[0093] Due to the different weights of the sensor detection results 350 and 360, that is, due to the increased weight of the sensor detection result 360 according to the above special association, during the regression process, the section 160 is refined by "attracting" it in the direction of the sensor detection result 360 with the increased weight. This is shown in Figure 4 C. The refinement of the section 160 in Figure 4 A can be recognized by comparing it with the refined section 420 as shown in Figure 4 C.

[0094] Due to the refinement of the section 160, the expansion of the sensor detection results 114 relative to the initial contour 140 is incorporated into the method for determining the contour. Due to the association and weighting processes described in the context of Figure 3 , the sensor detection results 114 that are closer to the distance sensor 110 and the vehicle 100 are set to have increased weights, which causes the refinement of the initial contour 140 to translate the contour in the direction towards the vehicle 100 or the sensor 110.

[0095] The effect of refining the initial contour 140 is stronger for the part of the initial contour 140 where the sensor detection results 114 have a large expansion relative to the initial contour 140 than for the part of the initial contour 140 where there is a small expansion or almost no expansion relative to the initial contour 140. This is shown in Figure 5The cases marked by B and C are illustrated. For both scenarios, the sensor detection results 114 in the middle region of the contour show a large expansion in two directions or dimensions (i.e., also in a direction perpendicular to the initial contour 140). Therefore, the sensor detection results 114 located closer to the vehicle 100 and the sensor 110 are set to have an increased weight, i.e., the farther sensor detection results 114 have a greater distance with respect to the vehicle 100 and the sensor 110 and are set to have a decreased weight.

[0096] As a result, in Figure 5 the two scenarios marked by B and C, the initial contour 140 is refined in such a way that the corresponding final contours 520, 530 are more strongly shifted in the direction towards the vehicle 100 compared to the initial contour 140 in the part where there is a large expansion of the sensor detection results 114. Therefore, the final contours 520, 530 represent the true contours of the corresponding objects more accurately than the initial contour 140. This can be recognized by comparison with the scenario marked by A, for which the association step, weighting step, and refinement step described above in the context of Figure 3 and Figure 4 are not applied to the initial contour 140.

[0097] Figure 6 FIG. 600 shows a flow chart illustrating a method for determining an object's contour with respect to a sensor. The sensor can be configured to provide a plurality of sensor detection results, and each sensor detection result can include the corresponding position where the object is located.

[0098] At 602, an initial contour of the object can be determined, the initial contour including a plurality of segments, each segment being related to a corresponding initial subset of the sensor detection results. At 604, corresponding surrounding groups in the sensor detection results can be associated with each segment of the initial contour. At 606, a corresponding weight can be provided to each of the sensor detection results in the corresponding surrounding group, each weight depending on the relative position of the sensor detection result with respect to the associated segment. At 608, each segment of the initial contour can be refined by using the weights of the sensor detection results in the corresponding surrounding group associated with the corresponding segment in order to determine a final contour including the refined segments for the object.

[0099] According to various embodiments, the initial contour may further include a plurality of vertices, and a segment of the initial contour may extend between a corresponding pair of vertices. One of the sensor detection results may be selected as the first vertex, and other vertices after the first vertex may be determined iteratively by selecting a corresponding initial subset of the sensor detection results relative to the corresponding previous vertex, and estimating the position of the next vertex by using the corresponding initial subset selected relative to the previous vertex.

[0100] According to various embodiments, selecting a corresponding initial subset of the sensor detection results for a previous vertex may include selecting sensor detection results located within a predefined region surrounding the previous vertex.

[0101] According to various embodiments, the predefined region may be provided as a rectangular bounding box, which may be centered on the corresponding vertex, and when iteratively determining a plurality of vertices, the rectangular bounding box may be translated from the corresponding one of the vertices to the subsequent vertex.

[0102] According to various embodiments, the sensor detection results selected for the initial subset of the previous vertex may be excluded from the initial subsets for subsequent vertices.

[0103] According to various embodiments, the sensor detection result selected as the first vertex may be confirmed by determining at least two sorted lists for the position of the sensor detection results, each sorted list referring to the corresponding coordinates of the reference position, selecting the sorted list having the largest difference between the first element and the last element in the list, and selecting the first element in the selected sorted list as the first vertex.

[0104] According to various embodiments, estimating the position of the next vertex by using the corresponding initial subset may include determining a segment vector extending from the previous vertex to the next vertex by calculating the geometric mean of the sensor detection results in the corresponding initial subset, wherein the geometric mean may provide the direction of the segment vector, and determining the farthest sensor detection result within the corresponding initial subset relative to the previous vertex, wherein the distance between the farthest sensor detection result and the previous vertex may define the absolute value of the segment vector.

[0105] According to various embodiments, corresponding main regions may be arranged symmetrically with respect to corresponding segments, and the sensor detection results of the corresponding surrounding group may include normal sensor detection results located within the main regions. A corresponding modified region may be determined by modifying the corresponding main region according to the distribution of the sensor detection results with respect to the sensors, and the sensor detection results of the corresponding surrounding group may further include special sensor detection results located within the modified region and outside the main region. The normal sensor detection results may be set to have a normal weight, and the special sensor detection results may be set to have an increased weight greater than the normal weight.

[0106] According to various embodiments, a corresponding modified region may be determined by translating the corresponding main region in a direction perpendicular to the corresponding segment, and the direction and amount of translation may depend on the distribution of the sensor detection results with respect to the sensors.

[0107] According to various embodiments, a corresponding segment vector may be determined as the difference between the position vectors of a pair of vertices associated with the corresponding segment, a corresponding normal vector perpendicular to the corresponding segment vector may be determined for each segment, and an auxiliary vector may be determined as the difference between the position vector of a reference near the sensor position and the position vector of the center of the segment closest to the sensor. For all segments, the amount of translation of the main region may be given by the dot product of the auxiliary vector and the normal vector of the segment closest to the sensor, and the direction of translation is given by the corresponding normal vector of the segment.

[0108] According to various embodiments, for each sensor detection result in the corresponding surrounding group, it may be determined whether the sensor detection result is located at the initial contour, on the sensor-facing inner side of the initial contour, or on the sensor-averted outer side of the initial contour. The sensor detection results located at the initial contour are set to have a normal weight, while the sensor detection results located on the inner side are set to have an increased weight greater than the normal weight, and the sensor detection results located on the outer side are set to have a decreased weight less than the normal weight.

[0109] According to various embodiments, refining each segment of the initial contour may include using a regression process to modify the corresponding segment to the weighted sensor detection results among the sensor detection results of the corresponding surrounding group associated with the corresponding segment.

[0110] Each of steps 602, 604, 606, 608 described above and other steps may be performed by computer hardware components.

[0111] Figure 7Shows a contour determination system 700 according to various embodiments. The contour determination system 700 may include an initial contour determination circuit 702, an association circuit 704, a weighting circuit 706, and a refinement circuit 708.

[0112] The initial contour determination circuit 702 may be configured to determine an initial contour of an object including a plurality of segments, each segment being related to a corresponding initial subset in the sensor detection results.

[0113] The association circuit 704 may be configured to associate a corresponding surrounding group of the sensor detection results with each segment of the initial contour.

[0114] The weighting circuit 706 may be configured to set a corresponding weight for each of the sensor detection results in the corresponding surrounding group, each weight depending on the relative position of the sensor detection results with respect to the associated segment.

[0115] The refinement circuit 708 may be configured to refine each segment of the initial contour by using the weights of the sensor detection results in the corresponding surrounding group associated with the corresponding segment, so as to determine a final contour including the refined segments for the object.

[0116] The initial contour determination circuit 702, the association circuit 704, the weighting circuit 706, and the refinement circuit 708 may be coupled to each other, for example, via an electrical connection 710 (such as a cable or a computer bus) or via any other suitable electrical connection to exchange electrical signals.

[0117] "Circuit" may be understood as any kind of logical implementation entity, which may be a dedicated circuit or a processor that executes a program stored in a memory, firmware, or any combination thereof.

[0118] Figure 8 Shows a computer system 800 having a plurality of computer hardware components configured to perform the steps of a computer-implemented method for predicting the corresponding trajectories of a plurality of road users according to various embodiments. The computer system 800 may include a processor 802, a memory 804, and a non-transitory data storage 806.

[0119] The processor 802 may execute instructions set in the memory 804. The non-transitory data storage 806 may store a computer program including instructions that may be transferred to the memory 804 and subsequently executed by the processor 802.

[0120] The processor 802, the memory 804, and the non-transitory data storage 806 may be coupled to each other, for example, via an electrical connection 710 (such as a cable or a computer bus) or via any other suitable electrical connection to exchange electrical signals.

[0121] As such, the processor 802 , the memory 804 , and the non-transitory data store 806 may represent the initial contour determination circuit 702 , the association circuit 704 , the weighting circuit 706 , and the refinement circuit 708 , as described above.

[0122] The terms “coupled” or “connected” are intended to include a direct “coupled” (eg, via a physical link) or direct “connected” as well as an indirect “coupled” or indirect “connected” (eg, through a logical link), respectively.

[0123] It should be understood that the above description of one of the methods may also be applicable to the contour determination system 700 and / or the computer system 800 .

[0124] Reference numerals list

[0125] 100 vehicles

[0126] 110 radar sensor

[0127] 112 Meter Filter View

[0128] 114 sensor detection results

[0129] 120 objects

[0130] 122 The true outline of the object

[0131] 130 Approximate outline according to related art

[0132] 140 polylines, initial outlines

[0133] 150 vertices

[0134] 151 First Vertex

[0135] 152 Second Vertex

[0136] 153 The third vertex

[0137] 154-156 Additional vertices

[0138] 160 sections

[0139] 161 Part 1

[0140] 165 The section closest to the vehicle

[0141] 200 Iteration process for determining the initial contour

[0142] 201-205 Iteration 1 to Iteration 5

[0143] 210 Selected sort direction

[0144] 221 First Square Bounding Box

[0145] 222 Second square bounding box

[0146] 223 - 225 Additional square bounding boxes

[0147] 232 Average point

[0148] 234 Furthest sensor detection result

[0149] 236 First segment vector

[0150] 238 Second segment vector

[0151] 310 Rectangular associated bounding box

[0152] 320 Translated associated bounding box

[0153] 330 Translation direction

[0154] 340 Auxiliary vector

[0155] 350 Sensor detection result of a normal association with normal weights

[0156] 360 Sensor detection result of a special association with increased weights

[0157] 370 Increased weight provided to the sensor detection result

[0158] 380 Decreased weight provided to the sensor detection result

[0159] 410 Distance from the detection result to the segment

[0160] 420 Refined segment

[0161] 520, 530 Final contour

[0162] 600 Flowchart illustrating a method for determining an object's contour with respect to sensors

[0163] 602 Step of determining an initial contour of an object including multiple segments, each segment being related to a corresponding initial subset in the sensor detection results

[0164] 604 Step of associating a corresponding surrounding group in the sensor detection results with each segment of the initial contour 606 Step of setting a corresponding weight for each in the sensor detection results in the corresponding surrounding group, each weight depending on the relative position of the sensor detection result with respect to the associated segment of the detection result

[0165] 608 Step of refining each segment of the initial contour by using the weights of the detection results in the corresponding surrounding group associated with the corresponding segment, in order to determine a final contour for the object including the refined segments

[0166] 700 Profile Determination System

[0167] 702 Initial Profile Determination Circuit

[0168] 704 Association Circuit

[0169] 706 Weighting Circuit

[0170] 708 Refinement Circuit

[0171] 710 Connection

[0172] 800 Computer System According to Various Embodiments

[0173] 802 Processor

[0174] 804 Memory

[0175] 806 Non-Temporary Data Storage

[0176] 808 Connection

Claims

1. A computer-implemented method for determining a contour (520, 530) of an object (120) with respect to a sensor (110), the sensor (110) being configured to provide a plurality of sensor detection results (114), each sensor detection result (114) including a corresponding position at the object (120). The method includes: determining an initial contour (140) of the object (120) including a plurality of segments (160), each segment (160) being related to a corresponding initial subset of the sensor detection results (114); associating a corresponding surrounding group of the sensor detection results (114) with each segment (160) of the initial contour (140); setting a corresponding weight for each of the sensor detection results (350, 350) in the corresponding surrounding group, each weight depending on the relative position (410) of the sensor detection result (114) with respect to the associated segment (160); and refining each segment (160) of the initial contour (140) by using the weights of the sensor detection results (114) in the corresponding surrounding group associated with the corresponding segment (160) to determine a final contour (520, 530) of the object (120) including the refined segments (420).

2. The method according to claim 1, wherein the initial contour (140) further includes a plurality of vertices (150); the segments (160) of the initial contour (140) extend between a corresponding pair of the vertices (150); selecting one of the sensor detection results (114) as a first vertex (151); iteratively determining additional vertices (152, 153, 154, 155, 156) after the first vertex (151) by selecting the corresponding initial subset of the sensor detection results (114) with respect to a corresponding previous vertex (151, 152, 153, 154, 155); and estimating the position of the next vertex (152, 153, 154, 155, 156) by utilizing the corresponding initial subset selected with respect to the previous vertex (151, 152, 153, 154, 155).

3. The method according to claim 2, wherein selecting the corresponding initial subset of the sensor detection results (114) for the previous vertex (151, 152, 153, 154, 155) includes selecting the sensor detection results (114) located within a predefined region (221, 222, 223, 224, 225) surrounding the previous vertex (151, 152, 153, 154, 155).

4. The method according to claim 3, wherein Provide the pre-defined region (221, 222, 223, 224, 225) as a rectangular bounding box centered at the corresponding vertices (151, 152, 153, 154, 155), and when iteratively determining multiple such vertices (150), translate the rectangular bounding box from a corresponding one of the vertices (151, 152, 153, 154, 155) to a subsequent vertex (152, 153, 154, 155, 156).

5. The method according to any one of claims 2 to 4, wherein exclude the sensor detection results (114) selected for the initial subset of the previous vertices (151, 152, 153, 154, 155) from the initial subset for the subsequent vertices (152, 153, 154, 155, 156).

6. The method according to any one of claims 2 to 5, wherein confirm the sensor detection result (114) selected as the first vertex (151) by determining at least two sorted lists for the position of the sensor detection result (114), each sorted list referring to the corresponding coordinates of the position; select the sorted list having the largest difference between the first element and the last element in the list, select the first element of the selected sorted list as the first vertex (151).

7. The method according to any one of claims 2 to 6, wherein estimating the position of the next vertex (152, 153, 154, 155, 156) by using the corresponding initial subset includes determining the segment vectors (236, 238) extending from the previous vertex (151, 152, 153, 154, 155) to the next vertex (152, 153, 154, 155, 156) by calculating the geometric mean of the sensor detection results (114) in the corresponding initial subset, wherein the geometric mean provides the direction of the segment vectors (236, 238), and determining the sensor detection result (234) farthest from the previous vertex (151, 152, 153, 154, 155) within the corresponding initial subset, wherein the distance between the farthest sensor detection result (234) and the previous vertex (151, 152, 153, 154, 155) defines the absolute value of the segment vectors (236, 238).

8. The method according to any one of claims 1 to 7, wherein arrange the corresponding main region (310) symmetrically with respect to the corresponding segment (160), the corresponding set of surrounding sensor detection results (114) includes the regular sensor detection results (350) located within the main region (310), determine the corresponding modified region (320) by modifying the corresponding main region (310) according to the distribution of the sensor detection results (114) relative to the sensors (110). The sensor detection results (114) corresponding to the surrounding group further include special sensor detection results (360), and the special sensor detection results (360) are located within the modified area (320) and outside the main area (310). Set the conventional sensor detection results (350) to have normal weights, and Set the special sensor detection results (360) to have an increased weight greater than the normal weight.

9. The method according to claim 8, wherein Determine the corresponding modified area (320) by translating the corresponding main area (310) in a direction perpendicular to the corresponding segment (160), and The direction and amount of translation depend on the distribution of the sensor detection results (114) relative to the sensor (110).

10. The method according to claim 9, wherein Determine the corresponding segment vectors (236, 238) as the difference in the position vectors of a pair of vertices (150) associated with the corresponding segment (160), For each segment (160), determine a corresponding normal vector perpendicular to the corresponding segment vector (236, 238), and determine the auxiliary vector (340) as the difference in the position vector of a reference near the sensor position and the position vector of the center of the segment (165) closest to the sensor (110). For all segments (160), the amount of translation of the main area (310) is given by the dot product of the auxiliary vector (340) and the normal vector of the segment (165) closest to the sensor (110), and the direction of translation is given by the corresponding normal vector of the segment (160).

11. The method according to any one of claims 1 to 10, wherein For each sensor detection result (114) in the corresponding surrounding group, determine whether the sensor detection result (114) is located at the initial contour (140), on the inner side of the initial contour (140) facing the sensor (110), or on the outer side of the initial contour (140) avoiding the sensor (110). Set the sensor detection results (114) located at the initial contour to have normal weights. Set the sensor detection results (114) located on the inner side to have an increased weight (370) greater than the normal weight, and Set the sensor detection results (114) located on the outer side to have a reduced weight (380) less than the normal weight.

12. The method according to any one of claims 1 to 10, wherein Refining each segment (160) of the initial contour (140) includes using a regression process to modify the corresponding segment (160) to the weighted sensor detection results (350, 360) in the sensor detection results (114) of the corresponding surrounding subset associated with the corresponding segment (160).

13. A computer system (800), the computer system (800) being configured to: Receiving a plurality of sensor detection results (114) from a sensor (110), the sensor (110) being configured to determine a corresponding position of the sensor detection results (114), Performing the computer-implemented method according to at least one of claims 1 to 12.

14. A vehicle (100), the vehicle (100) comprising: A sensor (110), the sensor (110) being configured to obtain a plurality of sensor detection results (114) and to determine a corresponding position of the sensor detection results (114), and The computer system (800) according to claim 13.

15. A non-transitory computer-readable medium comprising instructions for performing the computer-implemented method according to at least one of claims 1 to 12.

Images