Methods for automatic route assessment

Abstract

Method for evaluating the optical appearance of a driving route (28, 29, 30), wherein - from at least one motor vehicle (3) that is driving or has driven on the route (28, 29, 30), video data (6) containing an image sequence (13) of the route (28, 29, 30) which is transmitted to a computing device (4) via a respective communication link (7) and received - by an analysis device (12) on the basis of the video data (6) to images (14) of the image sequence (13) a correspondence value (15) concerning a degree of correspondence of the image (14) with a predetermined attractiveness criterion is determined and - the agreement values ​​(15) of the images (14) are combined to form an evaluation signal (10) for the image sequence (13) and - the evaluation signal (10) is provided as a description of the appearance of the route (28, 29, 30) and - a maximum range (16) of the evaluation signal (10) is detected in which at least one of the agreement values ​​(15) of the evaluation signal (10) is greater than a predetermined threshold (17), and - from the image sequence (13) at least one image (14) corresponding to the maximum range (16) of the image sequence (13) is determined and output, and - for each of the images (14) of the image sequence (13) a quality value (18) of an optical quality of the image (14) is generated based on a predetermined quality measure, wherein the quality measure specifies a signal-to-noise ratio and / or an image sharpness, and - the quality values ​​(18) are combined to form a quality signal (11) and - that at least one corresponding image (14) is determined from the maximum range (16) depending on the quality signal (11) and wherein - a playback duration of that part of the image sequence (13) for which the evaluation signal (10) has the maximum range (16) is determined and - depending on the playback duration either a video signal or only a single image (14) is output, whereby - for a route evaluation, each evaluation signal (10) is summarized by forming a mean value to obtain a single overall value (31) in order to obtain a compact statement about a journey route, and - the total value (31) for a journey and the extracted images are provided by the computing device (4) and - a navigation system (21) of a vehicle (5) and the computing unit (4) establish a communication link (22) and - the vehicle (5) obtains the extracted images and total values ​​from the computing unit (4) and graphically returns them to the driver as a route preview and route evaluation.

Description

[0001] The invention relates to a method for evaluating the visual appearance of a route. The evaluation determines whether the route or parts of the route, for example, are particularly scenic or feature architectural landmarks. The invention further develops a method known from US 2006 / 0271286 A1.

[0002] In the known method, motor vehicles capture video data with image sequences of routes along a road network. The video data is collected centrally on a server and displayed to a user who wishes to view information about the route. If a large number of motor vehicles generate video data, each route is filmed multiple times. To select from the video data, people must view and evaluate the image material. Only in this way is it possible to distinguish poor video recordings from good film recordings representative of the route. This makes the process complex.

[0003] WO 2014 / 054288 A1 discloses a method in which video data representing an image sequence of a route is generated by multiple motor vehicles. From the video data, the image sequence that shows the route without any vehicles traveling on it or with the smallest number of vehicles is selected. This image is then made available to drivers so that they can gain an impression of the road layout or course, even in heavy traffic.

[0004] US 2009 / 0171568 A1 discloses a method for displaying images based on the current location of a motor vehicle. This allows landmarks or prominent objects to be displayed to a driver as the driver approaches them. This can provide a driver with an indication of where to turn, for example, with their vehicle.

[0005] From the publication "Nagata, Daitchi et al.: SakuraSensor: A Participatory Sensing System for Detecting Flowering Cherries with Car-Mounted Smartphones" (HotMobile 2015, International Workshop on Mobile Computing Systems and Applications, February 12-13, 2015; URL: http: / / www.hotmobile.org / 2015 / papers / posters / Nagata.pdf), a method for detecting cherry blossoms is known. In this method, a large number of smartphones film the route taken by a motor vehicle. Software installed on the smartphone extracts short videos from the entire video recording and transmits them to a server upon request. This is intended to conserve the limited transmission capacity within the 3G / 4G communication network. The video data is then subjected to a histogram-based color analysis, with the color histogram used representing the color distribution of cherry blossoms.By comparing the video clips with the color histogram, it is determined whether blooming cherry trees or cherry blossoms are visible in the video clips. In conjunction with the smartphones' location data, this allows the route sections along which cherry blossoms can be observed to be identified.

[0006] Nowadays, drivers can get an idea of ​​various routes leading to their chosen destination before setting off. They can also view photos, videos, and reviews of routes online, for example.

[0007] To automate route selection, the image material, i.e., photos and videos, must have already been evaluated and categorized by other people. If the criteria are abstract, such as an assessment of "scenic beauty," then you are dependent on the judgment of another person. Otherwise, a search for a "scenic" route cannot be performed automatically.

[0008] A second problem lies in the evaluation of the volume of video data itself. If image sequences of the surroundings are collected centrally on a server from several vehicles, each with a video camera in each vehicle, the effort required to view this extensive video data is no longer economically justifiable.

[0009] The invention is based on the object of evaluating an optical appearance of driving routes in order to enable a systematic search for a driving route depending on its optical appearance.

[0010] The problem is solved by the subject matter of the independent patent claim. Advantageous developments of the invention result from the features of the dependent patent claims.

[0011] According to the invention, a method is provided for evaluating the visual appearance of a route. In this context, a route is, for example, a section of a road network, i.e., for example, a street or a stretch of road. In the manner already described, video data is received from at least one motor vehicle that is traveling or has traveled on the route. Preferably, several motor vehicles are provided. The video data from each motor vehicle contains an image sequence or a video of the route. The video data can, for example, be received by a central Internet server, i.e., a computing device connected to the Internet. On the basis of the video data for images of the image sequence, an analysis device determines to what degree, proportion, or percentage the image corresponds to a predetermined attractiveness criterion.For this purpose, a corresponding match value is determined, for example a value in a range from 0 to 100 percent. The attractiveness criterion can, for example, describe or define a "scenic" route or an "architecturally beautiful" route. The match value then indicates the extent or degree to which the route represented by the video data corresponds to the attractiveness criterion. For this purpose, for example, comparisons are made for individual images of the image sequence between the image and an optical property by which the attractiveness criterion is characterized. For example, green spaces, i.e. green image areas, can be representative of a scenically beautiful environment. The match value can then be determined depending on the proportion of a respective optical property (green image areas, image areas with building structures).

[0012] The analysis device generates a match value for several or all images of the image sequence. The match values ​​are combined according to the inventive method to form an evaluation signal for the image sequence. In other words, the match values ​​of the images represent amplitude values ​​of the evaluation signal. The evaluation signal is then provided as a description of the appearance of the route. It can, for example, be provided together with the video data or integrated into the video data. Using the method, several routes can of course also be provided with an evaluation signal in the manner described.

[0013] The invention provides the advantage that video data generated by motor vehicles, for example by means of video cameras during a journey, is automatically evaluated by a central computing device, for example an internet server, with regard to an attractiveness criterion. The image sequence of the video data is thus supplemented by an evaluation signal which indicates a match value or degree of affiliation with the attractiveness criterion for individual sections of the route. The evaluation signal thus provides a description of the appearance of the route for further automated processing. It is not necessary for a person to evaluate the video data. The aforementioned analysis device can, for example, comprise a program module of the computing device. The analysis device can, for example, be formed on the basis of a classifier or an algorithm for classification.

[0014] The invention has optional further developments, the features of which provide additional advantages.

[0015] According to a further development of the invention, the analysis device determines the match value of each image using a neural network. The neural network groups image elements of the image in a single-stage or multi-stage processing process. In other words, the image elements are summarized. This advantageously results in an abstraction or abstraction of the image elements into representative numbers or values. For example, green image elements in an image can be summarized to form a green region, which can then be represented by a single value that can, for example, represent the proportion of green space or foliage in the image. Using the neural network, as an alternative to evaluating a single image, it can also be provided to use the image and at least one other image of the image sequence as a basis for determining the match value.This advantageously results in a smoothing or stabilizing effect. Particularly preferred is the use of a so-called deep neural network, which in particular has more than three neural layers. This allows the image to be abstracted into a single matching value using only one neural network. No additional post-processing is then required to calculate the matching value.

[0016] According to a further development of the invention, training images are compared with each other and divided into groups using an unsupervised learning method to configure the neural network. In other words, the need for a person to select or mark the crucial image elements can advantageously be dispensed with when configuring the neural network. By comparing training images, it is automatically determined which image regions are important for distinguishing between the training images. For example, training images differ only slightly in the area of ​​the road section ahead, which is highly likely to be the same gray in all training images. The neural network is thus particularly sensitive or sensitized to the edge regions of the image, where colors and structures differ depending on the landscape characteristics.For example, one of the groups can have a grey border area for a journey through an urban area and a second group can represent a green border area, such as that which occurs when travelling through a countryside.

[0017] Algorithms for performing unsupervised learning of a neural network are known from the state of the art. However, grouping does not yet provide a way to assign a matching value to an image based on an attractiveness criterion using the neural network.

[0018] In a further step, example images, each of which has been assigned a predetermined match value, are assigned to one of the groups. This may involve a very small number of images that may have been rated by people. For example, people may determine that an example image contains particularly beautiful landscape elements. If this image is then assigned to one of the groups (which, for example, contains images with a green border area), it is then clear that images of the image sequence that are assigned to this group by the neural network are obviously classified by a person as having beautiful scenery. In contrast, an example image that was created during a trip through an urban area is not classified by the person as having beautiful scenery. If this example image is then assigned to another group (e.g.with a gray border area in the images), this image is assigned a lower match value than an image assigned to the group with a green border area. This further development results in the advantage that a large number of training images can be processed without evaluation by operators and only a relatively small number of example images compared to the training images are required to finally configure the neural network. The ratio of the number of example images to the number of training images can, for example, be less than 1 percent. This allows the analysis device to be configured in an automated process that only requires user intervention when determining the degree of match of the example images with the attractiveness criterion.

[0019] This embodiment is particularly advantageous in connection with an attractiveness criterion that relates to subjective assessments of aesthetic perception provided by at least one person. With a corresponding development, the example images are thus each provided with a subjective assessment of a route shown in the example image, provided by at least one person. An assessment is therefore made with regard to the attractiveness criterion. The assessment indicates, in particular, scenic beauty or architectural beauty or a degree of remoteness of the route. For example, the assessment can therefore state that the route shown in the example image is very scenic or not scenic. In particular, a classification can also be made according to a scale, i.e. a value for the subjective impression to be assessed or estimated.For example, the person can assign a value to the route for the attractiveness criterion “scenic” in a range from 0 to 100 percent.

[0020] According to the invention, an overall value for the route is calculated from the evaluation signal. This can, for example, be an arithmetic or geometric mean of the evaluation signal. The averaged evaluation signal can then be further divided into intervals, for example, to indicate a level of affiliation (e.g., 3 out of 5 stars) with the attractiveness criterion. Providing a single overall value for the entire route has the advantage of significantly reducing the computational effort required when comparing multiple routes.

[0021] According to the invention, a maximum range of the evaluation signal is also detected. In the maximum range, at least one of the match values ​​of the evaluation signal, i.e., an amplitude of the evaluation signal, is greater than a predetermined threshold value. At least one image of the image sequence corresponding to the maximum range is then determined and output from the image sequence. In other words, depending on the playback duration, either a video signal comprising multiple images or only a single image is output, wherein the images each have a match value greater than the threshold value according to the evaluation signal with the attractiveness criterion. Thus, a local maximum or the global maximum is determined in the evaluation signal, and at least one image from this maximum range is output. The user then receives at least one image from the image sequence that is particularly representative with regard to the attractiveness criterion.This has the advantage that a particularly meaningful image of the image sequence or a meaningful film excerpt from the image sequence is provided in an automated form.

[0022] According to the invention, a quality value for an optical quality of the image, determined based on a predetermined quality measure, is additionally generated for each of the images in the image sequence. The quality measure indicates a signal-to-noise ratio and / or image sharpness. The image sharpness and the signal-to-noise ratio can be determined in a conventional manner. A measure for occlusion by a vehicle traveling ahead can also be determined. Alternatively, a quality value can be determined for each of the images mentioned using a further neural network in the manner described, wherein training images and example images can again be used as a basis to enable unsupervised learning and subsequent configuration. The quality values ​​of the images are then combined to form a quality signal.Once a maximum range of the evaluation signal has been determined, at least one image from the maximum range can be additionally determined based on the quality signal. This offers the advantage that the at least one image corresponding to the maximum range is also selected based on image quality. This ensures that the output image has a predetermined minimum quality, for example, that a clear view of the surroundings is displayed.

[0023] The invention relates to the question of whether a single image or a video signal should be selected from the maximum range. For this purpose, the playback time of that part of the image sequence for which the evaluation signal has the maximum range is determined. For example, the number of images of the image sequence provided or available in the maximum range is determined. A playback time in relation to a normal playback speed can also be determined on the basis of the video data. Depending on the determined playback time, either the part of the image sequence, i.e. a video signal, or just an individual image from the part is output. This has the advantage that no undesirably short video signals are presented that cannot be mentally processed by a human viewer. In this case, it is more sensible to provide a still image or static single image that the person can view.

[0024] According to a further development of the invention, the evaluation signal is generated as a function of the time of day and / or the season and / or the weather. In other words, the match value is provided together with the information that this match was determined at a predetermined time of day and / or season and / or under predetermined weather conditions. For example, the attractiveness criterion "scenic beauty" may have been determined in summer or winter, resulting in different match values ​​of the images with the attractiveness criterion.

[0025] According to a development of the invention, an evaluation signal is generated for each of several routes in the manner described. In connection with a navigation assistance system, at least one route that leads to a predetermined navigation destination is determined. For each determined route, the user is given an indication regarding the attractiveness criterion based on the respective evaluation signal. A selection of one of the routes that the user enters, for example, can then be received. It can also be provided that the user selects the attractiveness criterion and that route is then selected for the navigation assistance which, according to its evaluation signal, best fulfills the attractiveness criterion, for example, has the highest overall value.The advantage of this training is that navigation assistance is provided which also takes into account subjective and / or emotional criteria, namely the attractiveness criterion.

[0026] The method according to the invention can be implemented by a computer for the Internet. This computer can thus realize the same advantages as the method.

[0027] An exemplary embodiment of the invention is described below. It shows: Fig. 1 is a schematic representation of a system by which an embodiment of the method according to the invention is carried out, Fig. 2 diagrams with a schematic progression of a rating signal and a quality signal, Fig. 3 a schematic representation of a navigation assistance, and Fig. 4 a flowchart of an embodiment of the method according to the invention.

[0028] The exemplary embodiment explained below is a preferred embodiment of the invention. In the exemplary embodiment, the described components of the embodiment each represent individual, independently considered features of the invention, which also further develop the invention independently of one another and are thus also to be considered as components of the invention, either individually or in a combination other than that shown. Furthermore, the described embodiment can also be supplemented by further features of the invention already described.

[0029] In the figures, functionally identical elements are provided with the same reference numerals.

[0030] Fig. 1 shows a system 1 formed by a motor vehicle fleet 2 with several motor vehicles 3, a central computing device 4, for example, an internet server, and another motor vehicle 5. The system 1 makes it possible to provide a driver of the motor vehicle 5 with information for different routes, indicating the extent to which these routes satisfy an abstract emotional attractiveness criterion, for example, the attractiveness criterion "scenic beauty" or "architectural beauty" or "remoteness." With the system 1, it is not necessary for the routes to be inspected by people. The system 1 is capable of evaluating the routes in an automated manner.

[0031] For this purpose, each motor vehicle 3 of the motor vehicle fleet 2 generates video data 6 while driving on one of the routes, which video data is transmitted to the computing device 4 via a respective communication connection 7. The communication connection 7 can be established, for example, via a mobile radio connection and / or an internet connection. The video data 6 can be generated by the motor vehicles 3 using a video camera 8. A detection area of ​​the video cameras 8 can be aligned in a front area of ​​the motor vehicle 3. The video data 6 thus each contain a video or an image sequence of images, each of which depicts an area surrounding the motor vehicle 3. The video data 6 can be transmitted via a radio module 9, for example a mobile radio module or a WLAN radio module (WLAN - wireless local area network), via the communication connection 7.

[0032] The video data 6 are processed by the computing device 4 such that an evaluation signal 10 and optionally a quality signal 11 are generated for each image sequence of the video data 6. Multiple evaluation signals 10 can also be generated for different attractiveness criteria. The evaluation signal 10 indicates the extent to which individual images of the image sequence of the video data 6 correspond to the abstract, emotional attractiveness criterion, for example, the attractiveness criterion "scenic beauty." "Extent" here means that a correspondence value is provided for individual images of the image sequence. The quality signal 11 can indicate an image quality for each image determined according to a quality criterion. The evaluation signal 10 and the optional quality signal 11 are generated by an analysis device 12 of the computing device 4.The analysis device 12 can, for example, be designed as a program module of the computing device 4.

[0033] Fig. Figure 2 illustrates how, for video data 6, the image sequence 13 from images 14 is used to depict a course of the respective journey of the motor vehicle 3 along the route over time t. Of the images 14 of the image sequence 13, Fig. 2 shows only a few of them. It shows how each image 14 is assigned a match value 15 by the analysis device 12, which indicates to what degree D or to what extent the image content of the image 14 corresponds to the attractiveness criterion. Fig. 2, an attractiveness criterion C1 is evaluated, which can be, for example, the attractiveness criterion "scenic beauty," "architectural beauty," or "remoteness." Each match value 15 can, for example, indicate the degree D of match with the attractiveness criterion C1 in a range from 0 percent to 100 percent. Combining the match values ​​15 results in the evaluation signal 10.

[0034] In Fig. 2 further illustrates a maximum range 16 of the evaluation signal 10. In this maximum range 16, the evaluation signal 10 has agreement values ​​15 that are greater than a predetermined threshold value 17. The threshold value 17 can, for example, represent the mean value of the evaluation signal 10 or the mean value weighted by a factor that can, for example, be in a range from 1.2 to 2.0.

[0035] The quality signal 11 can be formed from individual quality values ​​18, which can be determined for each of the images 14 by the analysis device 12. In Fig. 2 shows how images in the maximum range 16 exhibit a drop in quality 19, which may be caused, for example, by the camera 8 being obscured by a vehicle driving ahead.

[0036] In Fig. 1 further shows how the video data 6 with its evaluation signal 10 and its quality signal 11 can be stored in a memory 20.

[0037] In the motor vehicle 5, a driver can now be assisted in planning a trip by the computing device 4 using a navigation system 21. For this purpose, the navigation system 21 and the computing device 4 can establish a communication connection 22, which can be formed, for example, on the basis of a radio module 23 of the motor vehicle 5 and a communication module 24 of the computing device 4. The communication module 24 can be formed, for example, by a program module and / or a network connection device of the computing device 4. The radio module 23 can be configured in the same way as the radio modules 9.

[0038] Fig. 3 illustrates how the navigation system 21 or the computing device 4 can support route planning based on the evaluation signal 10 and the quality signal 11.

[0039] The navigation system 21 can have a display device 25, for example, a screen and / or a head-up display device (head-up display) of the motor vehicle. For the example, it is assumed that the driver wishes to travel from a starting position 26 of the motor vehicle 5 to a navigation destination 27. For example, three possible routes 28, 29, 30 have been determined by the navigation system 21. For each of the routes 28, 29, 30, the respective evaluation signal 10 or a total value 31 formed therefrom can be determined via the communication connection 22. For each route 28, 29, 30, for example, a total value 31 can then be determined, which can be calculated as an average of the respective evaluation signal 10. Fig. 3, total values ​​of 70 percent, 59 percent, and 10 percent were determined for routes 28, 29, and 30, respectively. The driver is thereby informed that with regard to the attractiveness criterion C1, for example, scenic beauty, route 28 has the highest total value 31, namely 70 percent.

[0040] For the further explanation of the example, it is assumed that the evaluation signal 10 consists of Fig. 2 belongs to the route 28. The computing device 4 and / or the navigation system 21 can now determine a position of the maximum area 16 along the route 28. Image information 32 relating to the maximum area 16 can be presented to the driver on the display device 25, for example a single image 14 or a part of the image sequence 13 formed from multiple images 14, i.e. a video excerpt. The image information 32 can also be determined as a function of the quality signal 11. For example, it can be taken into account that no images from the quality dip 19 are displayed. The driver thus receives image information 32 as to which, according to the attractiveness criterion C1, is the most impressive or important section of the route 28 and which visual impressions he can expect.

[0041] In Fig.4 shows an example of how the analysis device 12 can be configured to generate an evaluation signal 10 and a quality signal 11 for the video data 6 by means of the analysis device 12. In a step S1, a classifier, e.g. a neural network 33, can be trained on the basis of training images 34 using a method of unsupervised learning. The neural network 33 represents only one example of a trainable classifier. A support vector machine or another algorithm from the field of pattern recognition can also be used as the classifier. The training images 34 can, for example, be contained in training image sequences of driving routes that have been filmed from moving motor vehicles. Through the unsupervised learning in step S1, the training images 34 are divided into groups without an operator having to define the groups.Corresponding algorithms for the unsupervised learning of neural networks or other classifiers are known from the prior art. The training images 34 represent route sections that exhibit varying degrees of correspondence with a predetermined attractiveness criterion. For example, the training images 34 may have been generated during a city trip and a country trip.

[0042] In a step S2, the trained neural network 33 is now calibrated or adjusted such that the initially meaningless groups into which the training images 34 were divided using unsupervised learning are now assigned a respective degree of agreement or agreement value, wherein the agreement value represents the agreement with respect to a predetermined attractiveness criterion. For this purpose, example images 35 are determined, each with a predetermined agreement value 15. This can be determined, for example, by people. In step S2, it is determined which of the groups 35 the example images 35 are assigned to. Thus, each of the groups 35 can be assigned a corresponding agreement value 15.

[0043] In a step S4, a further neural network 37 or another classifier can be trained in a comparable manner based on training images 38. The training images 38 represent driving routes with different image qualities. The image quality can be defined, for example, by a signal-to-noise ratio and / or image sharpness and / or a degree of occlusion by vehicles driving ahead. In a step S5, the trained neural network 37 can be calibrated or configured in the manner described based on example images 39, each of which has been assigned a quality value 18. The quality values ​​18 of the example images 39 can also have been determined by people.

[0044] In a step S6, the analysis device 12 then processes an image sequence 13 from the video data 6 on the basis of the neural networks 33, 37. This results in an evaluation signal 10 and a quality signal 11 for the video data 6 in the manner described. This is achieved by assigning a match value 15 and a quality value 18 to each image of the image sequence 13. In a step S7, the generated signals can be stored in the memory 20. In a step S8, for example, the computing device 4 can determine the evaluation signal 10 for each of the routes 28, 29, 30 determined by a navigation system 21.

[0045] In a step S9, the navigation system 21 can select the route 28 for the navigation assistance that has the highest overall value 31 or that the driver selects after viewing the image information 32. This ensures that the route guidance leads along a route 28 that represents the most favorable or best route according to the attractiveness criterion C1.

[0046] The example shows how the computing device 4, based on machine learning, in particular using deep neural networks, enables image and video recordings to be automatically evaluated with regard to abstract emotional criteria, such as scenic beauty. The members of a vehicle fleet 2 or a swarm of vehicles continuously capture video recordings of the routes they travel and store the recordings on a server formed by the computing device 4. From the collected video recordings, values ​​for attractiveness criteria such as scenic beauty, architectural beauty, and remoteness are automatically extracted for each frame in the video signal using machine learning. This procedure results in a signal curve for the respective attractiveness criteria.For route evaluation, each evaluation signal can be summarized into a single overall value, for example by calculating an average, in order to obtain a compact statement about a route.

[0047] For local maxima of the evaluation signals for the attractiveness criteria, individual images or short video clips can be extracted at the corresponding position in the video sequences or image sequences. If a maximum is a short peak, a single image should be extracted; if a plateau-like peak is present, a short video clip should be extracted. A short peak can be defined as described above.

[0048] For all video recordings, the optical quality value of the image material should also be extracted using a machine learning method, as previously described. This signal should be used during the extraction of individual images to identify the most visually appealing individual image around a maximum range in the evaluation signal, and then extract precisely this individual image. If a driver wishes to obtain information about the various possible routes to a destination, their vehicle retrieves the extracted images and attractiveness criteria from the computing device 4 and returns them graphically to the driver as a route preview and route rating.

[0049] After a one-time setup, no human intervention is required to create the route evaluation and route preview. The video recordings can be recorded by vehicles during normal journeys. Since the required data can be recorded in the background, an evaluation and preview can be created over time for every route traveled by appropriately equipped vehicles. This automatically results in a very high level of route coverage in the medium term. The described process runs completely automatically and requires only limited resources and financial investment, including for the implementation of the machine learning itself.

[0050] The members of the vehicle swarm use cameras to record images of the route along individual journeys and send the video data to a central server via a radio module. Once sufficient video data has been collected on the server, the training of a machine learning process is initiated. The learning process, for example a deep neural network, must be able to achieve a sufficiently high level of abstraction to summarize image elements from each image to the extent that a match value for one of the described attractiveness criteria, such as scenic beauty, is summarized. During the training phase, the analysis engine used is to learn from training images without additional information, i.e., using unsupervised learning. Example images rated by humans for the respective attractiveness criterion are then made available to the analysis engine.Using these sample images, the analysis engine can assign a match score to the patterns abstracted during the training phase. After training and matching are complete, the analysis engine can evaluate all available video data, as well as any subsequent video data, against the specified attractiveness criteria.

[0051] In the same way, for example, another neural network should be trained with regard to image quality. All video recordings should also be evaluated with it, resulting in an additional quality signal 11. Each extracted attractiveness criterion thus results in an evaluation signal 10 and a quality signal 11 for each incoming video recording. A further algorithm identifies local maxima and / or minima in these signal curves. At the points in the video recordings that correspond to a local, short-term maximum in the signal curve of the evaluation signal, individual images are extracted for a short-term maximum and video clips for a broader or longer maximum. When extracting individual images, the individual image is taken from the point in the video signal where the image quality signal and the evaluation signal reach a common maximum. The extracted images are, for example,together with the averaged rating signals, i.e., generally the total value for a route, are provided on a server so that they can be retrieved by motor vehicles. When a driver searches for routes to a specific destination using their navigation system, the navigation system downloads extracted images, video clips, and rating signals via a radio module, preferably appropriate for the current time of day and year. The navigation system 21 onboard the motor vehicle then displays information graphically.

[0052] Overall, the example shows how the invention can provide a method for automatic route movement and creation of a route preview.

Claims

[1] Method for evaluating an optical appearance of a route (28, 29, 30), wherein - from at least one motor vehicle (3) which is traveling or has traveled on the route (28, 29, 30), video data (6) containing an image sequence (13) of the route (28, 29, 30) are received, which are transmitted to a computing device (4) via a respective communication connection (7), and - an analysis device (12) determines, on the basis of the video data (6), a respective match value (15) relating to a degree of match of the image (14) with a predetermined attractiveness criterion for images (14) of the image sequence (13), and - the match values ​​(15) of the images (14) are combined to form an evaluation signal (10) for the image sequence (13) and - the evaluation signal (10) is provided as a description of the appearance of the route (28, 29, 30) and - a maximum range (16) of the evaluation signal (10) is detected in which at least one of the match values ​​(15) of the evaluation signal (10) is greater than a predetermined threshold value (17), and - at least one image (14) of the image sequence (13) corresponding to the maximum range (16) is determined and output from the image sequence (13), and - for each of the images (14) of the image sequence (13), a quality value (18) of an optical quality of the image (14) is generated, which quality value is determined on the basis of a predetermined quality measure, wherein the quality measure indicates a signal-to-noise ratio and / or an image sharpness, and - the quality values ​​(18) are combined to form a quality signal (11) and - the at least one corresponding image (14) is determined from the maximum range (16) as a function of the quality signal (11) and wherein - a playback duration of that part of the image sequence (13) for which the evaluation signal (10) has the maximum range (16) is determined and - depending on the playback time, either a video signal or only a single image (14) is output, whereby - for a route evaluation, each evaluation signal (10) is combined into a single overall value (31) by forming an average in order to obtain a compact statement on a route, and - the total value (31) for a route and the extracted images are provided by the computing device (4) and - a navigation system (21) of a vehicle (5) and the computing device (4) establish a communication connection (22) and - the vehicle (5) obtains the extracted images and the total values ​​from the computing device (4) and returns them graphically to the driver as a route preview and route evaluation. [2] Method according to claim 1, wherein the analysis device (12) determines the match value (15) of each image (14) by means of a neural network (33) by which image elements of the image (14) or of the image (14) and at least one further image (14) of the image sequence (13) are combined. [3] Method according to claim 2, wherein, for configuring the neural network (33), on the one hand, training images (34) are compared with one another and divided into groups (35) by means of an unsupervised learning method and, on the other hand, example images (36) to which a predetermined match value (15) is assigned are assigned to the groups (35). [4] Method according to claim 3, wherein the example images (36) are each provided with a subjective assessment, provided by at least one person, of a route section shown in the example image (36), wherein the assessment indicates in particular a scenic beauty or an architectural beauty or a degree of remoteness of the route section. [5] Method according to one of the preceding claims, wherein a total value (31) for the route (28, 29, 30) is calculated from the evaluation signal (10). [6] Method according to one of the preceding claims, wherein the evaluation signal (10) is formed as a function of the time of day and / or the season and / or the weather. [7] Method according to one of the preceding claims, wherein an evaluation signal (10) is generated for each of a plurality of routes (28, 29, 30) and at least one route (28, 29, 30) leading to a predetermined navigation destination (27) is determined and a user is given an indication (31, 32) relating to the attractiveness criterion for each determined route (28, 29, 30) on the basis of the respective evaluation signal (10).

Images