Railway line detection device and program product

Abstract

Provided are a railway line detection device and a program that can more accurately detect a branch line of a railway. The railway line detection device of the present embodiment includes a line detection unit and a clustering unit. The line detection unit performs line detection of detecting one path of a railway line from an image obtained by photographing a traveling direction of a railway vehicle a plurality of times according to a number of detection parameters set in advance. The clustering unit detects a branch of the railway line in the image by classifying a plurality of line detection results of the line detection unit into a plurality of clusters.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to a railway line detection device and a program. BACKGROUND

[0002] There is an obstacle sensing system in front which monitors an obstacle that is likely to come into contact with a railway vehicle using an image captured by a camera that captures the front of the railway vehicle. In order to determine whether an object present in front is to be classified as an obstacle, a line along which the railway vehicle is scheduled to travel is detected from the captured image, and for this process, high precision is required. Since the braking distance of the railway vehicle until it stops is long, it is required to detect a line further away with high precision and at high speed. In addition, it is difficult to detect a branch line where a branch exists in the line with high precision.

[0003] PRIOR ART DOCUMENTS

[0004] PATENT DOCUMENTS

[0005] Patent Document 1: Japanese Patent Application Publication No. 2020-179798 SUMMARY

[0006] PROBLEMS TO BE SOLVED BY THE INVENTION

[0007] An object is to provide a railway line detection device and a program that can detect a branch line of a railway with higher precision.

[0008] MEANS FOR SOLVING THE PROBLEMS

[0009] The railway line detection device of the present embodiment has a line detection section and a clustering section. The line detection section performs line detection that detects one path of a railway vehicle from an image captured by a camera that captures the traveling direction of the railway vehicle a plurality of times according to the number of a plurality of detection parameters set in advance. The clustering section detects a branch of a line within the image by classifying a plurality of line detection results of the line detection section into a plurality of clusters. BRIEF DESCRIPTION OF DRAWINGS

[0010] Figure 1 is a block diagram showing one example of the functional structure of the railway line branch detection device of the first embodiment.

[0011] Figure 2 is a block diagram showing one example of the functional structure of the railway line branch detection device of the first embodiment.

[0012] Figure 3 is a flowchart showing one example of the operation of the railway line branch detection device of the first embodiment.

[0013] Figure 4is a view showing an example of visualizing a plurality of line detection results acquired by the line detection execution section of the first embodiment.

[0014] Figure 5 is a view showing an example of data configuration of the line detection result acquired by the line detection execution section of the first embodiment.

[0015] Figure 6 is a view showing an example of processing in which the line detection result clustering section of the first embodiment classifies a plurality of line detection results.

[0016] Figure 7 is a view showing an example of a method of clustering in the line detection result clustering section of the first embodiment.

[0017] Figure 8 is a view showing an example of a classification parameter in the line detection result clustering section of the first embodiment.

[0018] Figure 9 is a view showing an example of preprocessing in the line detection result clustering section of the first embodiment.

[0019] Figure 10 is a view showing an example of a method of clustering in the line detection result clustering section of the second embodiment.

[0020] Figure 11 is a view showing an example of a method of clustering in the line detection result clustering section of the third embodiment.

[0021] Figure 12 is a view showing an example of interpolation of a feature position in the line detection result clustering section of the fourth embodiment.

[0022] Explanation of Reference Numerals

[0023] 1 driving assistance device, 40 railway line branch detection device, 401 line detection execution section, 402 line detection execution parameter storage section, 403 line detection result clustering section, 404 line branch number acquisition section, 405 line database, 406 line detection result output section, 407 operation information database, CP10 to CP14 feature position, CPr0 to CPr4 feature position, I image, R line, RV railway vehicle, Wcp0 to Wcp2 offset width, Wr line width DETAILED DESCRIPTION

[0024] Embodiments of the present application will be described below with reference to the accompanying drawings. The embodiments are not intended to limit the present application. The drawings are schematic or conceptual, and the ratios of the parts, etc. are not necessarily the same as those in reality. In the description and drawings, the same elements as those described with respect to already appearing figures are designated with the same reference numerals, and detailed description is appropriately omitted.

[0025] <First Embodiment>

[0026] [Structure]

[0027] Figure 1 is a block diagram showing one example of the structure of the driving assistance device 1 of the first embodiment. The driving assistance device 1 is a system mounted on a railway vehicle RV traveling on a railway line R. The driving assistance device 1 detects an obstacle located in front of the railway vehicle RV, and notifies the driver of the sensing result of the obstacle. The driving assistance device 1, for example, is provided with an imaging device 10, a front obstacle detection device 20, an obstacle determination result notification device 30, a railway line branch detection device 40, a current position measuring device 50, and a GNSS (Global Navigation Satellite System) 60.

[0028] The imaging device 10 is composed of a camera that images the front of the railway vehicle RV, and outputs the imaged image to the front obstacle detection device 20 and the railway line branch detection device 40.

[0029] The GNSS 60 locates the current position of the railway vehicle RV based on the electric wave received from a satellite. The GNSS 60 outputs the located position information of the railway vehicle RV to the current position measuring device 50.

[0030] The current position measuring device 50 obtains the travel position of the railway vehicle RV based on the current position information obtained from the GNSS 60 and a line database. The current position measuring device 50 outputs the travel position of the railway vehicle RV to the railway line branch detection device 40.

[0031] The railway line branch detection device 40 detects the position of the line R existing in the image based on the image obtained from the imaging device 10 and the travel position obtained from the current position measuring device 50. The railway line branch detection device 40 detects the line R including a branch. The railway line branch detection device 40 outputs the detection result of the line R to the front obstacle detection device 20.

[0032] The front obstacle detection device 20 detects an obstacle present in the track R that is likely to come into contact with the running railway vehicle RV, based on the track position in the image acquired from the imaging device 10 and the image acquired from the railway track branch detection device 40. The front obstacle detection device 20 detects the obstacle, for example, by using image recognition using an AI (Artificial Intelligence) technique such as machine learning. However, the method of detecting the obstacle is not limited thereto. The front obstacle detection device 20 outputs the detection result of the obstacle to the obstacle determination result notification device 30.

[0033] The obstacle determination result notification device 30 refers to the information acquired from the front obstacle detection device 20, and notifies, for example, the driver attendant, the driving command station, the station attendant, and the passenger of the presence of the obstacle, and the like. The obstacle determination result notification device 30 can also notify, for example, a control device for automatic driving that controls the accelerator and the brake of the presence of the obstacle, and the like.

[0034] Figure 2 is a block diagram that shows one example of the functional structure of the railway track branch detection device 40 of the first embodiment.

[0035] The railway track branch detection device 40 includes a track detection execution section 401, a track detection execution parameter storage section 402, a track detection result clustering section 403, a track branch number acquisition section 404, a track database 405, a track detection result output section 406, and a running information database 407. Some or all of the track detection execution section 401, the track detection result clustering section 403, the track branch number acquisition section 404, and the track detection result output section 406 are implemented, for example, by a processor such as a CPU (Central Processing Unit) executing a program stored in a storage section (not shown). The storage section can be provided inside the railway track branch detection device 40 or outside the railway track branch detection device 40. In addition, some or all of these functional sections can be implemented by hardware such as an LSI (Large Scale Integration), an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array), or the like, have a circuit structure for implementing the functions of these functional sections, or can be implemented by a combination of software and hardware.

[0036] The line detection execution section 401 applies single (without assuming the existence of branches) line detection based on the image acquired from the image acquisition section 101 possessed by the imaging device 10. As for the line detection of the line detection execution section 401, line detection such as that of Japanese Patent No. 6776213 or Japanese Patent No. 6903477 corresponds to single line detection. The line detection execution section 401 performs multiple line detections based on the parameters (detection parameters) stored in the line detection execution parameter storage section 402, and acquires multiple line detection results. The line detection execution section 401 outputs the multiple line detection results to the line detection result clustering section 403.

[0037] That is, the line detection execution section (line detection section) 401 performs multiple line detections of a line R of a path of a railway vehicle RV from an image I obtained by imaging the traveling direction of the railway vehicle RV, according to the number of detection parameters set in advance. Details of the line detection will be described later with reference to Figure 4 and Figure 5 .

[0038] The line detection execution parameter storage section 402 is a database that stores a parameter group (parameter set) required for the line detection execution section 401 to operate.

[0039] The line branch number acquisition section 404 acquires the traveling position of the railway vehicle RV from the current position acquisition section 501 possessed by the current position measurement device 50, and acquires coordinate information of the line R being traveled from the line database 405. The line branch number acquisition section 404 acquires the number of branches of the line R included in the currently imaged image, and outputs it to the line detection result clustering section 403.

[0040] The line detection result clustering section 403 classifies (integrates) the multiple line detection results acquired from the line detection execution section 401 based on the number of branches acquired from the line branch number acquisition section 404.

[0041] In addition, the line detection result clustering section (clustering section) 403 detects branches of the line R within the image I by classifying the multiple line detection results of the line detection execution section 401 into multiple clusters. Details of the clustering will be described later with reference to Figures 6-9 .

[0042] The line detection execution section 401 acquires multiple line detection results that encompass all branch targets, but it is possible to obtain duplicate line detection results for each branch target. Therefore, the line detection result clustering section 403 acquires line detection results that exclude duplicates for each branch target, and outputs them to the line detection result output section 406.

[0043] The operation information database 407 is a database that stores information related to the route R of the scheduled travel of the railway vehicle RV. The operation information database 407 acquires the branch target of the scheduled travel based on the travel position of the railway vehicle acquired by the current position acquisition unit 501, and outputs it to the route detection result output unit 406.

[0044] The route detection result output unit 406 acquires the branch target of the scheduled travel from the operation information database 407 and the detection result of the route R for all the branch targets from the route detection result clustering unit 403, and outputs the route detection result for the branch target of the scheduled travel to the object collision determination unit 201 possessed by the forward obstacle detection device 20.

[0045] That is, the route detection result output unit 406 outputs at least one route detection result. In more detail, the route detection result output unit 406 outputs at least one route detection result in the cluster corresponding to the scheduled travel path of the railway vehicle RV based on the position of the railway vehicle RV. In addition, the route detection result clustering unit 403 causes the route detection result output unit 406 to output the route detection result closest to the center of the classified cluster. The center of the cluster is sometimes also referred to as the centroid (center of gravity).

[0046] [PROCESS FLOW]

[0047] Figure 3 is a flowchart showing one example of the operation of the railway route branch detection device 40 of the first embodiment.

[0048] In step S001, the route detection execution unit 401 acquires the image obtained by photographing the front of the railway vehicle RV from the photographing device 10.

[0049] In steps S002 and S003, the route detection execution unit 401 acquires a plurality of route detection results based on the parameters (detection parameters) stored in the route detection execution parameter storage unit 402 using the image obtained by photographing the front of the railway vehicle RV acquired from the photographing device 10 as input. The route detection execution unit 401 performs a plurality of route detections by changing the parameter group.

[0050] In step S0004, the route detection result clustering unit 403 classifies the plurality of route detection results acquired in step S003, and excludes the route detection results that are duplicated for each branch target. The route detection result clustering unit 403 can also use the number of route branches included in the photographed range from the current vehicle position, which can be acquired from the route branch number acquisition unit 404, when classifying the plurality of route detection results. The route detection result clustering unit 403 can also use the information of the route width in the image when classifying.

[0051] In step S005, the operation information database 407 obtains the predetermined branch target based on the travel position of the railway vehicle obtained from the current position acquisition unit 501, and outputs it to the track detection result output unit 406. The track detection result output unit 406 obtains the track detection result of the predetermined branch target based on the track detection results classified by the track detection result clustering unit 403, and outputs it to the forward obstacle detection device 20.

[0052] In step S006, the obstacle detection device 20 detects obstacles that have the potential to collide with the line detection results obtained from the line detection result output unit 406.

[0053] [Single Line Test]

[0054] Figure 4 This is a diagram illustrating an example of visualizing multiple line detection results obtained by the line detection execution unit 401 of the first embodiment.

[0055] Image I acquired from imaging device 10 contains branch lines that go straight and turn right. The line detection execution unit 401 can detect a single line; therefore, by performing multiple (three times in this figure) line detections based on parameters stored in the line detection execution parameter storage unit 402, results such as detection results A to C can be obtained. Detection results A and B represent the line detection results for the straight line R. On the other hand, detection result C represents the line detection result for the right-turning line R. Regarding detection results A and B, it can be seen that repeated line detection results were obtained for the straight line R.

[0056] like Figure 4 As shown, the line detection execution unit 401 determines the characteristic positions CP10-CP14 and CPr0-CPr4 respectively for the detection results A to C. The lines obtained by connecting the determined characteristic positions CPr0-CPr4 and the lines obtained by connecting the characteristic positions CP10-CP14 are respectively derived as the right line R1 and the left line R2 included in the line R.

[0057] Figure 5 This is a diagram illustrating an example of the data structure of the line detection results obtained by the line detection execution unit 401 of the first embodiment.

[0058] The line detection execution unit 401 represents the line detection results as a list of points at characteristic locations and outputs it to the line detection result clustering unit 403. Additionally, in Figure 5 In the image, the coordinates of four points are shown for detection results A through C. However, for... Figure 4The shown detection results A to C actually have coordinate values of 10 points of feature positions CPl0 to CPl4, CPr0 to CPr4, respectively.

[0059] The detection result A and the detection result B are both line detection results for a straight line. However, between the detection result A and the detection result B, the coordinates of the feature positions are not necessarily the same values. This is due to, for example, the line detection being performed with different detection parameters.

[0060] [Method of single line detection]

[0061] The line detection execution section 401 performs single line detection in Japanese Patent No. 6903477, for example, by changing the detection parameters. The parameters stored in the line detection execution parameter storage section 402 are, for example, a parameter group including the search range and the search angle (angle θ) in Japanese Patent No. 6903477. In this case, in a certain parameter group, a different search range or a different search angle is stored compared to other parameter groups. By changing the parameter group used as the detection parameter, the single line detected sometimes changes. In addition, the detection parameter changes according to the algorithm of the single line detection method employed.

[0062] [Clustering]

[0063] Figure 6 is a diagram showing one example of the processing by the line detection result clustering section 403 of the first embodiment to classify a plurality of line detection results.

[0064] As shown in Figure 4 , the line detection execution section 401 obtained repeated results for a straight line. Therefore, the line detection result clustering section 403 classifies similar line detection results and excludes the repeated line detection results. In the example shown in Figure 6 , the detection result A is repeated with the detection result B, so it is excluded, and the detection results B and C are displayed in an overlapping manner. The line detection result clustering section 403 can also import processing to define the distance between the detected lines and integrate according to the distance (refer to Figure 8 ). The line detection result clustering section 403 can also apply an existing clustering process such as the k-means method or the Mean-Shift method.

[0065] [Method of clustering]

[0066] Figure 7 is a diagram showing one example of the method of clustering in the line detection result clustering section 403 of the first embodiment. Figure 7An example in which feature positions CPlO to CPl2, CPrO to CPr2 are determined within the image I is shown. In addition, the Y coordinates of the nth points (n = 1 to 3) from the front in the feature positions CPlO to CPl2 and the feature positions CPrO to CPr2 are substantially the same.

[0067] The line detection result clustering section 403 classifies a plurality of line detection results into a plurality of clusters based on position information of the line detection results on the image I. The position information of the line detection results on the image I is, for example, pixel positions of the feature positions CPlO to CPl2, CPrO to CPr2. The line detection result clustering section 403 performs clustering using, for example, Euclidean distances of the pixel positions.

[0068] The line detection result clustering section 403 classifies a plurality of line detection results into a plurality of clusters by calculating distances (for example, Euclidean distances) between a plurality of line detection results for each part of the divided position information (pixel positions).

[0069] In the example shown in FIG. 6, the feature positions CPlO to CPl2, CPrO to CPr2 of six points are shown for one line detection result. The line detection result clustering section 403 performs clustering for each of the points. Figure 7

[0070] The line detection result clustering section 403 performs clustering, for example, by sequentially extracting the feature position CPl2, the feature position CPl 1, the feature position CPlO, the feature position CPr2, the feature position CPr 1, and the feature position CPrO. Thus, in the example shown in FIG. 6, the line detection result clustering section 403 performs clustering six times based on two (1 x 2) dimensional data of the X coordinates and the Y coordinates of one feature position. Figure 7

[0071] In addition, the order of the feature positions CPlO to CPl2, CPrO to CPr2 in which clustering is performed is not limited to the order described above.

[0072] [Classification parameters for clustering]

[0073] The classification parameters for clustering differ depending on the algorithm of the clustering selected. Hereinafter, a case in which the number of branches and the line width are used as the classification parameters is described. The algorithm of the clustering using the number of branches is, for example, the k-means method. The algorithm of the clustering using the line width is, for example, the mean shift method.

[0074] [Clustering based on the number of branches]

[0075] ​​The line detection result clustering section 403 classifies the plurality of line detection results into a number of clusters corresponding to the number of branches of the line R in the predetermined travel path of the railway vehicle RV, based on the position of the railway vehicle RV. As described with reference to Figure 2 As described, the number of branches of the line R is acquired from the line database 405 by the line branch number acquisition section 404. In a case where the number of branches in the vicinity of the current position of the railway vehicle RV is known in advance, the number of branches can also be used as an input to clustering. Thereby, the number of clusters classified within the image I can be determined in advance. As a result, the accuracy of clustering can be improved.

[0076] In Figure 4 In the example shown, the number of branches of the line R is one. In this case, the line detection result clustering section 403 classifies the plurality of line detection results (detection results A to C) into two clusters. In one cluster, the line R of the detection results A and B extends straight. In the other cluster, the line R of the detection result C extends straight until halfway, like the detection results A and B, and then extends curvedly to the right.

[0077] In addition, the line detection result clustering analysis section 403 can also stop the classification of the line detection results in a case where the number of branches based on the position of the railway vehicle RV is zero. Thereby, it is possible to achieve that branch detection is not performed at a position where there is no branch.

[0078] [Clustering based on line width]

[0079] Figure 8 is a diagram showing one example of a classification parameter in the line detection result clustering section 403 of the first embodiment.

[0080] The line detection result clustering section 403 classifies the plurality of line detection results into a number of clusters based on a classification parameter corresponding to the line width Wr. The classification parameter is the distance between clusters.

[0081] In Figure 8 In the example shown, two line detection results are shown. Between the two line detection results, the offset width (Euclidean distance) Wcp0 of the feature position CP10 and the offset width Wcp1 of the feature position CP11 are smaller than the line width Wr. On the other hand, between the two line detection results, the offset width Wcp2 of the feature position CP12 is larger than the line width Wr.

[0082] In a case where there are branches of the line R, i.e., two paths of clusters, generally, a large deviation occurs in the feature positions between the two line detection results. As a threshold value of the deviation of the feature positions, for example, the line width Wr can be used. Thus, in Figure 8In the example shown, the line detection result clustering section 403 classifies two line detection results into two clusters. As a result, the line detection result clustering section 403 detects the branch of the line R.

[0083] In addition, the line detection result clustering section 403 may, for example, also use, as a classification parameter, a value obtained by multiplying the line width by a prescribed coefficient, in accordance with a clustering method or the like.

[0084] [Preprocessing of Clustering]

[0085] Figure 9 is a diagram showing one example of preprocessing in the line detection result clustering section 403 of the first embodiment.

[0086] In the image I, generally, the line width of the line R at a position far from the railway vehicle RV is shown smaller than the line width of the line R at a position close to the railway vehicle RV. In this case, it can be difficult to perform branch detection of the line R at a position far from the railway vehicle RV. This is because, on the image I, the shift width of the feature position on the far side (for example, the shift width Wcp2 shown in Figure 8 is shown smaller than the shift width of the feature position on the near side (for example, the shift width Wcp0 shown in Figure 8 ). That is, the result of clustering can greatly depend on the shift width of the feature position on the near side.

[0087] Therefore, the line detection result clustering section 403 performs correction processing of the detection result of the line R before performing clustering. For example, the line detection result is corrected in advance using a captured result of the line R in a straight section. In the straight section, the line R extends in a substantially straight line with substantially the same line width from the near side to the far side. The line detection result clustering section 403 corrects the feature positions CPio- CPi2, CPr0- CPr2 so that the line width is substantially constant from the near side to the far side. Thereby, the branch of the line R at a position far from the railway vehicle RV on the image I can be detected with higher accuracy.

[0088] That is, the line detection result clustering section 403 corrects the position information (pixel position) of the line detection result based on the line width. The line detection result clustering section 403 classifies a plurality of line detection results, of which the position information has been corrected, into a plurality of clusters. In more detail, the line detection result clustering section 403 corrects the position information (pixel position) of the line detection result so that the line width is normalized between the near side of the image I and the far side of the image I.

[0089] As described above, according to the first embodiment, the line detection execution unit 401 performs multiple line detections based on a predetermined number of multiple detection parameters, detecting a line R of a path for the railway vehicle RV from an image I obtained by capturing images of the RV's direction of travel. The line detection result clustering unit 403 detects branches of line R within image I by classifying the multiple line detection results into multiple clusters. This allows for more accurate detection of branches of line R.

[0090] Line detection is typically more difficult when branches are present. Other methods for detecting line branches include using machine learning to detect the presence or absence of switches. However, this method requires preparing images of the branches (switches) for machine learning. In other words, it requires training data, which can increase import costs.

[0091] In contrast, the first embodiment can extend the detection of a single line and detect branches of line R based on the line detection results. In this case, no learning data is required for machine learning, thus suppressing the increase in import costs.

[0092] Furthermore, in the line detection of Japanese Patent No. 6903477, the Y-coordinates of the nth point (n=1 to 3) from the nearest point between feature positions CP10~CP12 and feature positions CPr0~CPr2 are approximately the same. Depending on the clustering algorithm, clustering can also be performed using the X-coordinate instead of the Y-coordinate. In this case, the line detection result clustering unit 403 performs six clustering operations based on the X-coordinate of a feature position using one (1×1)-dimensional data. For example, the line detection result clustering unit 403 performs six one-dimensional Euclidean distance calculations.

[0093] Alternatively, the line detection execution unit 401 can also perform line detection using other line detection methods that detect a single line.

[0094] In addition, the number of line test results is not limited to Figure 4 The three are shown. The more line detection results for an image I, the better it is at removing off-values, which is preferable.

[0095] <Second Implementation Method>

[0096] Figure 10 This diagram illustrates an example of the clustering method in the line detection result clustering unit 403 of the second embodiment. In the second embodiment, the clustering method differs from that in the first embodiment.

[0097] The line detection result clustering unit 403 clusters each of the left and right paired feature positions of the right line R1 and the left line R2.

[0098] The line detection result clustering section 403, for example, extracts a pair of the feature positions CP12 and CPr2 to cluster. Next, the line detection result clustering section 403, for example, extracts a pair of the feature positions CP11 and CPr1 to cluster. Next, the line detection result clustering section 403, for example, extracts a pair of the feature positions CP10 and CPr0 to cluster. Thus, in the example shown in FIG. 8, the line detection result clustering section 403 performs clustering three times on four (2 x 2) dimensional data of the X coordinates and the Y coordinates of two feature positions. Figure 10 In the example shown, the line detection result clustering section 403 performs clustering three times on four (2 x 2) dimensional data of the X coordinates and the Y coordinates of two feature positions. The line detection result clustering section 403, for example, performs the operation of the four-dimensional Euclidean distance three times.

[0099] In addition, in the line detection of Japanese Patent No. 6903477, the Y coordinates of the nth point (n = 1 to 3) from the front between the feature positions CP10 to CP12 and the feature positions CPr0 to CPr2 are substantially the same. According to the algorithm of clustering, clustering can also be performed using the X coordinates without using the Y coordinates. In this case, the line detection result clustering section 403 performs clustering three times on two (2 x 1) dimensional data of the X coordinates of two feature positions. The line detection result clustering section 403, for example, performs the operation of the two-dimensional Euclidean distance three times.

[0100] The other structures of the railway line branch detection device 40 of the second embodiment are the same as the corresponding structures of the railway line branch detection device 40 of the first embodiment, and thus detailed description thereof is omitted.

[0101] As in the second embodiment, the method of clustering can also be changed. The railway line branch detection device 40 of the second embodiment is able to obtain the same effects as the first embodiment.

[0102] <Third Embodiment>

[0103] Figure 11 is a diagram showing one example of the method of clustering in the line detection result clustering section 403 of the third embodiment. In the third embodiment, the method of clustering is different compared to the first embodiment.

[0104] The line detection result clustering section 403 classifies a plurality of line detection results into a plurality of clusters by performing the operation of the distance (for example, the Euclidean distance) between a plurality of line detection results together for all position information (pixel positions).

[0105] The line detection result clustering section 403, for example, extracts all of the feature positions CP10 to CP12 and CPr0 to CPr2 to cluster. Thus, in the example shown in FIG. 8, the line detection result clustering section 403 performs clustering three times on four (2 x 2) dimensional data of the X coordinates and the Y coordinates of two feature positions. Figure 11In the illustrated example, the line detection result clustering unit 403 performs clustering of twelve (6 x 2) dimensions based on the X coordinates and the Y coordinates of the six feature positions once. The line detection result clustering unit 403, for example, performs an operation of the twelve-dimensional Euclidean distance once.

[0106] In addition, in the line detection of Japanese Patent No. 6903477, the Y coordinates of the nth points (n = 1 to 3) from the front between the feature positions CPio to CPi2 and the feature positions CPr0 to CPr2 are substantially the same. According to the algorithm of clustering, clustering using the X coordinates can be performed instead of using the Y coordinates. In this case, the line detection result clustering unit 403 performs clustering of six (6 x 1) dimensional data based on the X coordinates of the six feature positions once. The line detection result clustering unit 403, for example, performs an operation of the six-dimensional Euclidean distance once.

[0107] The other structures of the railway line branch detection device 40 of the third embodiment are the same as the corresponding structures of the railway line branch detection device 40 of the first embodiment, and thus detailed description thereof is omitted.

[0108] As in the third embodiment, the method of clustering can be changed. The railway line branch detection device 40 of the third embodiment can obtain the same effects as the first embodiment.

[0109] <Fourth Embodiment>

[0110] Figure 12 FIG. 7 is a diagram illustrating one example of interpolation of the feature positions CPio to CPi2 and CPr0 to CPr2 in the line detection result clustering unit 403 of the fourth embodiment. The fourth embodiment differs from the first embodiment in that clustering is performed by interpolating the feature positions CPio to CPi2 and CPr0 to CPr2.

[0111] In the line detection of Japanese Patent No. 6903477, the Y coordinates of the nth points (n = 1 to 3) from the front between the feature positions CPio to CPi2 and the feature positions CPr0 to CPr2 are substantially the same. However, for example, depending on the method of line detection, there are cases in which the values of the Y coordinates greatly differ. In this case, the line detection result clustering unit 403 classifies a plurality of line detection results into a plurality of clusters based on interpolation position information obtained by interpolating between a plurality of position information (coordinate positions). The line detection result clustering unit 403, for example, performs spline interpolation to interpolate the coordinates of the feature positions.

[0112] In the line detection of Japanese Patent No. 6903477, the Y coordinates of the nth points (n = 1 to 3) from the front between the feature positions CPio to CPi2 and the feature positions CPr0 to CPr2 are substantially the same. However, for example, depending on the method of line detection, there are cases in which the values of the Y coordinates greatly differ. In this case, the line detection result clustering unit 403 classifies a plurality of line detection results into a plurality of clusters based on interpolation position information obtained by interpolating between a plurality of position information (coordinate positions). The line detection result clustering unit 403, for example, performs spline interpolation to interpolate the coordinates of the feature positions. Figure 12In the illustrated example, the line detection result clustering unit 403 interpolates the characteristic positions CPio to CPi2, and calculates the interpolation position information of the left line R2. Similarly, the line detection result clustering unit 403 interpolates the characteristic positions CPr0 to CPr2, and calculates the interpolation position information of the right line Rl. The line detection result clustering unit 403, for example, sets 0, 1, 2,..., Y' as the Y coordinates to calculate the X coordinates of the interpolation position information of the left line R2. Note that Y' is a predetermined integer. The line detection result clustering unit 403 sets 0, 1, 2,..., Y' as the Y coordinates to calculate the X coordinates of the interpolation position information of the right line Rl.

[0113] The line detection result clustering unit 403 clusters using Y'+l pieces of the interpolation position information of the left line R2 and the right line Rl as the characteristic positions.

[0114] The other structures of the railway line branch detection device 40 of the fourth embodiment are the same as the corresponding structures of the railway line branch detection device 40 of the first embodiment, and thus detailed description thereof is omitted.

[0115] As in the fourth embodiment, clustering can be performed by interpolating the characteristic positions CPio to CPi2 and CPr0 to CPr2. The railway line branch detection device 40 of the fourth embodiment can achieve the same effects as the first embodiment. In addition, the second embodiment or the third embodiment can be combined in the railway line branch detection device 40 of the fourth embodiment. That is, the method of clustering can be changed.

[0116] Although several embodiments of the present application have been described, these embodiments are presented by way of example and are not intended to limit the scope of the application. These embodiments can be implemented in various other ways, and various omissions, substitutions, and changes can be made without departing from the spirit of the application. These embodiments and modifications thereof are included within the scope and spirit of the application, and are also included within the scope of the application and equivalents thereof recited in the claims.

Claims

1. A railway track branch detection device, characterized by, Possessing: a line detection section that performs a plurality of times line detection of a line of a path of a railway vehicle from an image obtained by photographing a traveling direction of the railway vehicle, acquires a plurality of line detection results, and a clustering section that detects a branch of the line within the image by classifying the plurality of line detection results of the line detection section into a plurality of clusters through clustering processing.

2. The railway line branch detection device according to claim 1, characterized in that the clustering section classifies a plurality of the line detection results into a number of clusters corresponding to a number of branches of the line in a predetermined traveling path of the railway vehicle based on a position of the railway vehicle.

3. The railway line branch detection device according to claim 2, characterized in that the clustering section stops classification of the line detection results in a case where the number of branches based on the position of the railway vehicle is zero.

4. The railway line branch detection device according to claim 1, characterized in that the clustering section classifies a plurality of the line detection results into a plurality of clusters based on a classification parameter corresponding to a line width.

5. The railway line branch detection device according to claim 1, characterized in that the clustering section classifies a plurality of the line detection results into a plurality of clusters based on position information of the line detection results on the image.

6. The railway line branch detection device according to claim 5, characterized in that the clustering section classifies a plurality of the line detection results, for which the position information has been corrected based on a line width, into a plurality of clusters.

7. The railway line branch detection device according to claim 6, characterized in that the clustering section corrects the position information of the line detection results so that the line width is normalized between a near side of the image and a far side of the image.

8. The railway line branch detection device according to claim 5, characterized in that the clustering section classifies a plurality of the line detection results into a plurality of clusters based on interpolation position information obtained by interpolating between a plurality of the position information.

9. The railway line branch detection device according to claim 5, characterized in that the clustering section classifies a plurality of the line detection results into a plurality of clusters by calculating a distance between a plurality of the line detection results for each part of the position information that has been divided.

10. The railway line branch detection device according to claim 5, characterized in that the clustering section classifies a plurality of the line detection results into a plurality of clusters by calculating a distance between a plurality of the line detection results for all of the position information.

11. The railway line branch detection device according to claim 1, characterized by further comprising a line detection result output section that outputs at least one of the line detection results.

12. The railway line branch detection device according to claim 11, characterized in that ​ ​ The line detection result output section outputs at least one of the line detection results in the cluster corresponding to a predetermined travel path of the railway vehicle, based on the position of the railway vehicle.

13. The railway line branch detection device according to claim 11, wherein The clustering section causes the line detection result output section to output the line detection result closest to the center of the classified cluster.

14. A program product having a program, characterized by The program causes the computer to execute a plurality of times line detection of a line of a path of a railway vehicle from an image taken of a travel direction of the railway vehicle, based on a number of detection parameters set in advance, to obtain a plurality of line detection results, The program causes the computer to detect a branch of a line within the image by clustering processing to classify the plurality of line detection results into a plurality of clusters.

Images