Systems and Methods for Automatic Semantic Map Generation

Abstract

The present invention relates to a system and method for automatic semantic map generation, which at least includes a processing system having at least one processing device. The processing system is configured to generate a probabilistic road model for road elements. The processing system is configured to generate a probabilistic traffic model for the road model. The processing system is configured to perform statistical inference on the road model and the traffic model. The processing system is configured to generate an estimate of a semantic road model based on the statistical inference. The processing system is configured to generate a semantic map including the estimate of the semantic road model.

Description

Technical Field

[0001] The present disclosure generally relates to maps, and more particularly to automatic semantic map generation. Background Art

[0002] Generally, some maps include measurable information about the environment. For example, these maps can include the position and shape of lane markings and other features in the environment. In addition to providing an indication of the roads in a geographic area, these maps can also include some driving information for these roads. However, these maps are typically handcrafted to include this driving information, resulting in a labor-intensive and time-consuming process, as well as some inaccuracies on the map. Summary of the Invention

[0003] The following is an overview of certain embodiments described in detail hereinafter. The presentation of the described aspects is merely to provide a brief overview of these certain embodiments to the reader, and the description of these aspects is not intended to limit the scope of the present disclosure. Indeed, the present disclosure may cover various aspects that may not be explicitly set forth hereinafter.

[0004] In an example embodiment, a method includes: generating a probabilistic road model via a processing system having at least one processing device based on road elements associated with geometric data corresponding to a geographic area. The method includes generating a probabilistic traffic model for the road model. The method includes calculating a statistical inference result via the processing system based on the road model and the traffic model. The method includes generating an estimate of a semantic road model via the processing system based on the statistical inference result.

[0005] In an example embodiment, a system includes at least one computer-readable medium that stores geometric data, traffic data, and computer-readable data. The system includes a processing system communicatively coupled to the at least one computer-readable medium. The processing system includes at least one processing device. The system is configured to execute the computer-readable data to implement a method that includes generating a probabilistic road model for road elements based on the geometric data. The method includes generating a probabilistic traffic model for the road model. The method includes calculating a statistical inference result based on the road model and the traffic model. The method includes generating an estimate of a semantic road model based on the statistical inference result. The method includes generating a semantic map that includes the estimate of the semantic road model.

[0006] In an example embodiment, a non-transitory computer-readable medium includes computer-readable data that, when executed by a processing system having at least one processing device, is configured to implement a method. The method includes generating a probabilistic road model for road elements. The method includes generating a probabilistic traffic model for the road model. The method includes calculating a statistical inference result based on the road model and the traffic model. The method includes generating an estimate of a semantic road model based on the statistical inference result.

[0007] These and other features, aspects, and advantages of the present invention are further elucidated by the following detailed description of certain exemplary embodiments with reference to the accompanying drawings, in which like characters or reference numerals denote like parts throughout the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 is a diagram of a system according to an example embodiment of the present disclosure.

[0009] Figure 2 is a flowchart of a method according to an example embodiment of the present disclosure.

[0010] Figure 3A is a non-limiting example of a map fragment for a geographic region according to an example embodiment of the present disclosure.

[0011] Figure 3B is according to an example embodiment of the present disclosure for Figure 3A a non-limiting example of a semantic map fragment with lane type classification for a geographic region associated with

[0012] Figure 3C is according to an example embodiment of the present disclosure for Figure 3A another non-limiting example of a semantic map fragment with lane type classification for a geographic region associated with

[0013] Figure 4A is a non-limiting example of a map fragment for a geographic region according to an example embodiment of the present disclosure.

[0014] Figure 4B is according to an example embodiment of the present disclosure for Figure 4A a non-limiting example of a semantic map fragment with lane segmentation for a geographic region associated with

[0015] Figure 4C is according to an example embodiment of the present disclosure for Figure 4A another non-limiting example of a semantic map fragment with lane segmentation for a geographic region associated with

[0016] Figure 5Ais a non - limiting example of a map fragment for a geographical area according to an exemplary embodiment of the present disclosure.

[0017] Figure 5B is a non - limiting example of a semantic map fragment with a lane merging relationship for a geographical area associated with Figure 5A according to an exemplary embodiment of the present disclosure.

[0018] Figure 5C is a non - limiting example of a semantic map fragment with a lane merging relationship for a geographical area associated with Figure 5A according to an exemplary embodiment of the present disclosure.

[0019] Figure 6A is a non - limiting example of a map fragment for a geographical area according to an exemplary embodiment of the present disclosure.

[0020] Figure 6B is a non - limiting example of a semantic map fragment with a traffic light association for a geographical area associated with Figure 6A according to an exemplary embodiment of the present disclosure.

[0021] Figure 6C is a non - limiting example of a semantic map fragment with a traffic light association for a geographical area associated with Figure 6A according to an exemplary embodiment of the present disclosure. Detailed Description

[0022] From the foregoing description, it will be understood that the above - described embodiments and many of their advantages have been shown and described by way of example, and it will be apparent that various changes may be made in the form, construction, and arrangement of the components without departing from the disclosed subject matter or sacrificing one or more of its advantages. Indeed, the descriptive form of these embodiments is merely explanatory. These embodiments allow for various modifications and alternative forms, and the appended claims are intended to cover and include such changes rather than being limited to the particular forms disclosed, but rather are intended to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the present disclosure.

[0023] Figure 1 is an illustration of system 100 according to an exemplary embodiment. In the exemplary embodiment, system 100 includes at least a semantic map generator 120, which is configured to communicate with at least one non - transitory computer - readable medium 110. In the exemplary embodiment, computer - readable medium 110 includes one or more electrical, electronic, or computer hardware storage devices or any combination thereof. Non - limiting examples of computer - readable medium 110 include any suitable storage device, memory circuit, random - access memory (RAM), read - only memory (ROM), any computer disk, any type of memory hardware, or any combination thereof. In Figure 1In [the example embodiment], the semantic map generator 120 may be communicatively coupled to computer-readable medium 110A and computer-readable medium 110B. In an example embodiment, computer-readable medium 110A and computer-readable medium 110B are different computer memory entities or different memory portions of the same computer memory entity. For example, computer-readable medium 110 may include computer-readable medium 110A and computer-readable medium 110B, which are different computer memory entities or different memory portions of the same computer memory entity.

[0024] In an example embodiment, as discussed herein, computer-readable medium 110A includes various relevant data related to generating, constructing, or providing a semantic map and / or a semantic road model. In this regard, for example, the various relevant data includes data collected via one or more data collection processes 10 for semantic map generation. Additionally or alternatively, the various relevant data includes data that is not collected via one or more data collection processes 10 for semantic map generation, but is related to generating a semantic road model and / or an estimation of a semantic map.

[0025] In an example embodiment, data collection process 10 includes obtaining sensor data from various sensors and storing the sensor data on computer-readable medium 110A. In this regard, for example, the various sensors include at least one camera, lidar, radar, GPS, odometry, any suitable sensor, or any combination thereof. In an example embodiment, the sensor data is processed at a central location and transmitted to computer-readable medium 110A. In an example embodiment, the sensor data includes geometric data. In an example embodiment, the sensor data includes at least measurable information related to a selected geographic region and / or environment.

[0026] In an example embodiment, data collection process 10 includes various methods for obtaining and collecting data related to generating a semantic road model and / or a semantic map. For example, data collection process 10 includes collecting and recording sensor data via various sensors associated with at least one dedicated instrumented test vehicle or fleet. In an example embodiment, each vehicle is equipped with various sensors and driven along a pre-planned route in traffic, thereby facilitating the initial task of estimating road elements from geometric data. In this regard, data collection process 10 advantageously takes into account the driving trajectories of the vehicles and the sensor measurements from the various sensors. Additionally, by including the pre-planned route and / or driving trajectory, system 100 is enabled to obtain and process relevant data, which avoids some of the ambiguities that may otherwise occur when estimating lanes from only geometric data (e.g., lane markings).

[0027] In an example embodiment, data collection process 10 includes one or more highly automated driving vehicles. In an example embodiment, data collection process 10 includes one or more vehicles having an advanced driver assistance system (e.g., a highway assistance system) such that each vehicle is configured to maintain a log of sensor data and transmit the log of sensor data to system 100 either via a wireless connection (such as an LTE connection) or during maintenance via a diagnostic port, via a computer-readable medium 110. Additionally or alternatively, data collection process 10 includes one or more human-driven vehicles that are configured to generate and / or maintain a log of sensor data. As another example, data collection process 10 includes obtaining traffic data that is collected via one or more mobile phone applications that use the built-in location services of each mobile phone to collect at least vehicle location and speed information. In this regard, as illustrated by one or more of the non-limiting examples above, data collection process 10 is advantageous in obtaining a target data set that is particularly relevant and useful for generating an effective and accurate estimate of a semantic road model and / or a semantic map.

[0028] In an example embodiment, semantic map generator 120 is a processing system that includes at least one electrical, electronic, and / or computer processing device. For example, in Figure 2 , semantic map generator 120 includes one or more computer servers and / or computer terminals that at least include a simultaneous localization and mapping (SLAM) module 130, a semantic map hypothesis generation module 140, and a verification module 150. In an example embodiment, each module of semantic map generator 120 includes hardware, software, or a combination of hardware and software. In an example embodiment, one or more modules of semantic map generator 120 are remote, local, or a combination thereof relative to each other and / or relative to semantic map application 20. Additionally, semantic map generator 120 is not limited to such a module configuration as shown in Figure 1 , but other embodiments include configurations having more modules, fewer modules, other functional modules, other logic units, or any suitable combination thereof, as long as semantic map generator 120 is configured to implement one or more of the objectives described herein.

[0029] In an example embodiment, the SLAM module 130 is configured to communicate with the computer-readable medium 110A. More specifically, the SLAM module 130 is configured to obtain various relevant data from the computer-readable medium 110A. In an example embodiment, after obtaining the various relevant data, the SLAM module 130 is configured to generate road elements, localization features, maps, or any combination thereof by implementing at least one process involving SLAM. In this regard, for example, the SLAM module 130 is configured to generate a map that includes lane markings, signs, road boundaries, other SLAM elements, or any combination thereof. Additionally, the SLAM module 130 is configured to communicate with another module (not shown), such as a localization module that is to provide localization features, the localization features including data related to the current position of the body within the geographical area.

[0030] In an example embodiment, the semantic map hypothesis generation module 140 is configured to communicate with the SLAM module 130. In this regard, the semantic map hypothesis generation module 140 is configured to obtain and / or receive at least geometric data, traffic data, map data, other relevant data, or any combination thereof from the SLAM module 130 and / or the computer-readable medium 110A. In an example embodiment, the semantic map hypothesis generation module 140 at least includes statistical inference software having an application programming interface (API), a domain-specific language (DSL), any suitable technology, or any combination thereof. In an example embodiment, the semantic map hypothesis generation module 140 is configured to generate semantic map element hypotheses. More specifically, in an example embodiment, the semantic map hypothesis generation module 140 is configured to generate a probabilistic road model based at least on road elements, map data, geometric data, any relevant data, or any combination thereof. Additionally, the semantic map hypothesis generation module 140 is configured to generate at least one probabilistic traffic model based at least on the probabilistic road model, traffic data, any suitable data, or any combination thereof. Additionally, the semantic map hypothesis generation module 140 is configured to generate at least one estimate of a semantic road model based at least on the probabilistic road model, the probabilistic traffic model, any suitable data, or any combination thereof. In an example embodiment, each estimate of the semantic road model represents a feasible interpretation of the various relevant data (e.g., geometric data, traffic data, etc.) associated with the geographical area. Additionally, in an example embodiment, the semantic map hypothesis generation module 140 is configured to communicate with the verification module 150.

[0031] In an example embodiment, the verification module 150 is configured to evaluate the validity of each estimate of the semantic road model. In this regard, the verification module 150 is configured to select an appropriate hypothesis (e.g., maximum likelihood, hypothesis testing, etc.) and / or determine whether each estimate of the semantic road model is considered valid. In this regard, for example, the verification module 150 is configured to make this validity determination by evaluating, for example, the result of a comparison with a predetermined criterion. In an example embodiment, for example, the verification module 150 is configured to perform a comparison between traffic prediction data for an estimate of the semantic road model and traffic data from a traffic log of actual traffic for at least one corresponding road segment in the geographic region. In such a case, the verification module 150 is configured to determine that the estimate of the semantic road model is invalid when the correlation between the traffic prediction data and the actual traffic data is insufficient. Alternatively, the verification module 150 is configured to determine that the estimate of the semantic road model is valid when the correlation between the traffic prediction data and the actual traffic data is sufficient.

[0032] In an example embodiment, when it is determined that an estimate of the semantic road model is considered invalid, the verification module 150 is configured to identify such road elements and / or semantic data that have insufficient correlation and / or a relatively high degree of ambiguity with respect to road element type, relationship classification, etc. After this information has been identified, the verification module 150 is configured to request additional data, the goal of which is to obtain this information for further data collection. In this regard, the verification module 150 is advantageously configured to identify regions with insufficient information and provide feedback to the system 100 regarding, for example, identifying or estimating additional data that is necessary to meet data requirements and / or a more reliable estimate of the semantic road model.

[0033] In an example embodiment, when it is determined that an estimate of the semantic road model is considered valid, the verification module 150 is configured to generate a semantic map including the estimate of the semantic road model. In an example embodiment, the generation of the semantic map includes updating an existing map to include the semantic road model or creating a new map including the semantic road model. In an example embodiment, the verification module 150 is further configured to store at least the semantic map (and any corresponding data) in at least one computer-readable medium 110B. In an example embodiment, the computer-readable medium 110B can be accessed via at least one other system, the at least one other system including hardware, software, or a combination thereof, and providing one or more applications related to the semantic map. For example, as Figure 1 shown, the computer-readable medium 110B is configured to communicate with another system that includes the semantic map application 20. In an example embodiment, the semantic map application 20 provides autonomous route planning, navigation features, positioning, any suitable map-related functions, or any combination thereof.

[0034] Figure 2 FIG. 2 is a flowchart that illustrates an example of a method implemented by system 100. In an example embodiment, the method may be implemented by any suitable hardware technology, software technology, or any combination of hardware and software technologies. Additionally, the method is advantageous in generating one or more semantic maps that provide various insights regarding one or more map elements. In an example embodiment, the semantic map includes geometric data, positioning data, semantic data, other relevant data, or any combination thereof. In this regard, the automatic generation of semantic maps is advantageous in various applications including, for example, navigation, route planning, positioning, other applications, or any combination thereof. Additionally, the automatic generation of semantic maps is advantageous for various entities such as humans, robots, mobile machines, autonomous vehicles, highly autonomous vehicles, and the like.

[0035] In an example embodiment, at step 202, system 100 generates a probabilistic road model. In this regard, for example, system 100 obtains various relevant data associated with a geographic region from computer-readable medium 110A to generate a probabilistic road model. For example, system 100 is configured to collect geometric data regarding a geographic region and provide a probabilistic road model based on the geometric data. In an example embodiment, the probabilistic road model provides a structure for identifying the presence of possible road elements or geographic elements based on geometric data. Non-limiting examples of road elements or geographic elements include lanes, traffic lights, stop lines, yield points, crosswalks, and the like. Additionally or alternatively, the probabilistic road model indicates how these identified elements are related to each other. For example, in a non-limiting example, based on geometric data, the probabilistic road model is configured to provide an indication of adjacent lanes and an indication of lane change transitions. Additionally, based on geometric data, the probabilistic road model is configured to provide an indication of lane order and the manner in which these lanes are sequentially arranged with any applicable continuation transitions. As another example, based on geometric data, the probabilistic road model is configured to provide an indication of merging or intersecting lanes and the priority of each lane among / within a set of lanes in the same merge or intersection.

[0036] In an example embodiment, at step 204, system 100 generates a probabilistic traffic model for the road model. In an example embodiment, the probabilistic traffic model provides the likelihood of traffic behavior observed for the road model generated at step 202. In this regard, for example, the probabilistic traffic model indicates the likelihood of traffic behavior given a particular arrangement of the road model and / or semantic road features and relationships.

[0037] In an example embodiment, at step 206, system 100 generates at least one estimate of a semantic road model. In an example embodiment, system 100 is configured to perform statistical inference, which involves computing, simulation, analysis, machine learning, any suitable processing, or any combination thereof. Additionally, in an example embodiment, the statistical inference is performed at least by software including an application programming interface (API), a domain-specific language (DSL), any suitable technology, or any combination thereof. In an example embodiment, the semantic map hypothesis generation module 140 performs various processes and / or computations for Bayesian inference, statistical hypothesis testing, any suitable probability / statistical method, or any combination thereof. In this regard, by performing statistical inference (e.g., hypothesis testing), the semantic map hypothesis generation module 140 is configured to manage complexity by considering a set of associations and relationships with a reasonable probability of existence.

[0038] In an example embodiment, for instance, after obtaining a probabilistic road model and a probabilistic traffic model, system 100 performs processes and / or computations for Bayesian inference to provide statistical inference results based on an estimate of the semantic road model given the observed geometric and traffic data. For example, in a non-limiting scenario, if traffic data shows that traffic vehicles never or rarely drive between two lane markings after a statistically large number of observations, system 100 is configured to determine that the area between the two lane markings has a lower probability of being a driving lane. Based on the traffic data, system 100 is configured to determine and / or generate output data indicating that the area is more likely to be classified as a non-driving shoulder or a median strip than as a driving lane. Similarly, if traffic data shows that traffic in the first lane at an uncontrolled intersection almost never yields to the second lane, system 100 is configured to determine and / or generate output data indicating that the first lane is very likely to have priority over the second lane at least based on traffic rules or local customs. Additionally, system 100 is configured to utilize geometric data (such as a stop sign) and relationship data (such as the association of the stop sign with the second lane) to determine whether there is a greater likelihood of the existence of the priority relationship (i.e., the first lane has priority over the second lane).

[0039] In an example embodiment, at step 208, system 100 determines whether the estimate of the semantic road model is valid. In an example embodiment, system 100 evaluates the validity of the estimate of the semantic road model based on a number of predefined criteria. In this regard, for example, the predefined criteria include one or more probabilities, likelihood values, confidence levels, thresholds, any statistically significant values, or any combination thereof. More specifically, in an example embodiment, for example, after receiving the estimate of the semantic road model from the semantic map hypothesis generation module 140, the verification module 150 is configured to compare the traffic prediction data for the estimate of the semantic road model with the actual traffic data from the traffic log for the section of the road, and to evaluate the result of this comparison using the predefined criteria to determine whether the estimate of the semantic road model is valid. Additionally or alternatively, the verification module 150 is configured to perform one or more other verification calculations to determine whether the estimate of the semantic road model meets one or more predefined criteria.

[0040] In an example embodiment, at step 210, system 100 generates a semantic map including the estimate of the semantic road model. In this regard, for example, the semantic map generator 120 (e.g., via the semantic map hypothesis generation module 140 and / or the verification module 150) generates the semantic map after verifying the estimate of the semantic road model at step 208. In an example embodiment, the semantic map is a representation of at least one geographical area that includes at least semantic data, positioning data, map data, geometric data, other relevant data, or any combination thereof. In an example embodiment, the semantic map is advantageous in providing a valid representation of the geographical area with information related to the identification of the various elements of the geographical area and the relationships between these various elements. Additionally, the semantic map is configured to be output via an I / O device such as a display device. Additionally or alternatively, the semantic map is configured to be connected to or integrated with the semantic map application 20, which is operable to provide route planning, positioning, mapping, any relevant applications, or any combination thereof.

[0041] In an example embodiment, at step 212, system 100 is configured to request and / or obtain additional data for generating the estimate of the semantic road model and / or the semantic map. For example, as Figures 1 to 2 shown, system 100 is configured to request additional data when it determines that the estimate of the semantic road model is invalid. By requesting and obtaining additional data, system 100 is configured to provide a more reliable estimate of the semantic road model in subsequent iterations. More specifically, after obtaining the additional data, system 100 is configured to proceed to step 202 to iterate through the method 200 again, as Figure 2As shown, another estimate of the semantic road is generated based on the additional data. In this way, using the additional data, system 100 is configured to generate an estimate of the semantic road model, which system 100 believes to be valid with a greater likelihood than a previous estimate of the semantic road model.

[0042] As described above, method 200 includes: assuming a probabilistic road model for how road elements relate to each other, and assuming a probabilistic traffic model for how traffic participants interact based on the probabilistic road model. Additionally, method 200 includes generating an estimate of the semantic road model and using the recorded traffic data to infer which relationships exist and which are active for each road element of the semantic road model. In this regard, method 200 is advantageous in constructing a semantic map using semantic data, which is applicable to various applications.

[0043] Figures 3A to 3C Illustrates a non-limiting example of various map fragments related to a geographic region. More specifically, Figure 3A Illustrates map fragment 300, which is generated by SLAM module 130 and includes geometric data. In this case, the geometric data includes a first lane marking 302 and a second lane marking 304. Although map fragment 300 indicates the presence of the first lane marking 302 and the second lane marking 304, map fragment 300 does not provide any meaningful information about the first lane marking 302 and the second lane marking 304.

[0044] Figure 3B Illustrates a first semantic map fragment 310 associated with the same geographic region as the geographic region of map fragment 300 that Figure 3A is followed. In an example embodiment, the first semantic map fragment 310 is generated by semantic map generator 120. In this regard, system 100 is configured to generate the first semantic map fragment 310, for example, after validating an estimate of the semantic road model based on traffic data, which indicates that there are typically two traffic flows in the corresponding part of the geographic region. Additionally, as Figure 3B shown, the first semantic map fragment 310 provides meaningful information related to the geometric data and / or road elements. In this regard, for example, the first semantic map fragment 310 includes a first lane identifier 312 to identify the space between the first lane marking 302 and the second lane marking 304 as a driving lane. Additionally, the first semantic map fragment 310 includes a second lane identifier 314 to identify the space on the opposite side of the second lane marking 304 as a driving lane. Additionally, in an example embodiment, the semantic data of the first semantic map fragment 310 includes the driving direction for the two driving lanes. In an example embodiment, system 100 is configured to determine the driving direction based on traffic data.

[0045] Figure 3C illustrates a second semantic map fragment 320 associated with a geographic region that is the same as the geographic region of the map fragment 300 with respect to Figure 3A . In an example embodiment, the second semantic map fragment 320 is generated by the semantic map generator 120. In this regard, the system 100 is configured to generate the second semantic map fragment 320, for example, after validating an estimate of the semantic road model based on traffic data, which indicates that there is typically a traffic flow and occasional vehicle movement in the corresponding part of the geographic region. Additionally, as Figure 3C shown, the second semantic map fragment 320 provides meaningful information related to geometric data and / or road elements. In this regard, for example, the second semantic map fragment 320 includes a first lane identifier 312 to identify the space between the first lane marking 302 and the second lane marking 304 as a driving lane. Additionally, the second semantic map fragment 320 includes a shoulder identifier 322 to identify the space on the opposite side of the second lane marking 304 as a road shoulder or a non-driving shoulder. That is, in contrast to the first semantic map fragment 310, the second semantic map fragment 320 identifies the space as a shoulder rather than a driving lane. Additionally, in the example embodiment, the semantic data of the second semantic map fragment 320 further includes driving directions for the first driving lane and the shoulder.

[0046] As Figure 3B and 3C shown, there are various possible interpretations for the geometric data and / or road elements presented in the map fragment 300. For example, the first semantic map fragment 310 indicates the presence of two driving lanes, while the second semantic map fragment 320 indicates the presence of one driving lane and one shoulder. These different interpretations convey or elaborate different driving conventions, rules, and habits, because a driving lane is a driving option, while a shoulder is more of an emergency option rather than a driving option. In the absence of such information, the map fragment 300 itself cannot provide sufficient information for an entity to plan and utilize the road elements according to driving rules and habits. In contrast, each of the first semantic map fragment 310 and the second semantic map fragment 320 provides insights that enable an entity to plan and utilize at least the road elements according to driving rules and habits.

[0047] Figures 4A to 4C illustrates a non-limiting example of various map fragments related to a geographic region. More specifically, Figure 4A illustrates a map fragment 400, which is generated by the SLAM module 130 and includes geometric data. In this case, the geometric data includes a first lane marking 402, a second lane marking 404, a third lane marking 406, and a fourth lane marking 408. Additionally, as Figures 4A to 4CAs shown in each of them, the second lane marking 404 includes a section 404A in which there is a discontinuity in the second lane marking 404 or an interruption with little lane marking. Additionally, as Figures 4A to 4C As shown in each of them, the third lane marking 406 includes a section 406A in which there is a discontinuity in the third lane marking 406 or an interruption with little lane marking. Although the ground image segment 400 here indicates the presence of the first lane marking 402, the second lane marking 404, the third lane marking 406, and the fourth lane marking 408, the ground image segment 400 does not provide any meaningful information regarding the meaning associated with the first lane marking 402, the second lane marking 404, the third lane marking 406, the fourth lane marking 408, the section 404A, and the section 406A.

[0048] Figure 4B illustrates a first semantic ground image segment 410 associated with a geographical area that is the same as the geographical area of the ground image segment 400 associated with Figure 4A In an example embodiment, the first semantic ground image segment 410 is generated by the semantic map generator 120. In this regard, the system 100 is configured to generate, for example, the first semantic ground image segment 410 after validating the estimation of the semantic road model based on traffic data, which indicates that in the corresponding part of the geographical area, there is generally one traffic flow in the first direction 418A and two traffic flows in the second direction 418B. Additionally, as Figure 4B As shown in it, the first semantic ground image segment 410 provides meaningful information related to geometric data and / or road elements. In this regard, for example, the first semantic ground image segment 410 includes a first lane identifier 412 to identify the space between the first lane marking 402 and the second lane marking 404 as a driving lane. Additionally, the first semantic ground image segment 410 includes a second lane identifier 414 to identify the space between the third lane marking 406 and the fourth lane marking 408 as a driving lane. Additionally, between the first lane identifier 412 and the second lane identifier 414, the first ground image segment includes a third lane identifier 416 to identify the space between the second lane marking 404 and the third lane marking 406 as a driving lane.

[0049] Furthermore, in an example embodiment, the first semantic ground image segment 410 also includes the driving directions for these driving lanes. More specifically, as Figure 4B As shown in it, the first driving lane associated with the first lane identifier 412 includes the first direction 418A. Meanwhile, as Figure 4BAs shown, the second and third driving lanes associated with the second and third lane identifiers 414 and 416 include a second direction 418B opposite to the first direction 418A. In this regard, for various semantic map applications 20, such as highly automated driving applications, autonomous driving applications, navigation applications, and other applications, including driving directions is highly advantageous.

[0050] Figure 4C Illustrates a second semantic map fragment 420 associated with a geographic area that is the same as the geographic area of the map fragment 400 of the ground Figure 4A In an example embodiment, the second semantic map fragment 420 is generated by the semantic map generator 120. In this regard, the system 100 is configured to generate the second semantic map fragment 420, for example, after verifying the estimate of the semantic road model based on traffic data, which indicates that in the corresponding part of the geographic area, there are generally two traffic flows that merge into a single traffic flow in the first direction 418A, and there are generally two other traffic flows that merge into a single traffic flow in the second direction 418B. In addition, as Figure 4C As shown, the second semantic map fragment 420 provides meaningful information related to geometric data and / or road elements. The second semantic map fragment 420 includes a first lane identifier 412 to identify the space between the first lane marking 402 and the second lane marking 404 as a driving lane. Additionally, the second map fragment 420 includes a second lane identifier 414 to identify the space between the third lane marking 406 and the fourth lane marking 408 as a driving lane. Additionally, in contrast to the first semantic map fragment 410, for the space between the second lane marking 404 and the third lane marking 406, the second semantic map fragment 420 includes a third lane identifier 422A for a part of the space and a fourth lane identifier 422B for another part of the space. In this regard, compared with the segments 404A and 406A of the first map fragment 410, the second semantic map fragment 420 assigns different meanings to the segments 404A and 406A related to the discontinuity or interruption in the second lane marking 404 and the third lane marking 406. More specifically, as Figure 4C As shown, the second semantic map fragment 420 identifies the geometric data of the map fragment 400 as including a first lane 412 in the first direction 418A and a second lane 414 in the second direction 418B, but also includes an interpretation that there is a merge between the first lane 412 and the fourth lane 422B and a merge between the second lane 414 and the third lane 422A. In this regard, the first lane 412 and the fourth lane 422B are associated with driving mainly in the first direction 418A, while the second lane 414 and the third lane 422A are associated with driving mainly in the second direction 418B.

[0051] AsFigure 4B and 4C As shown in 4C , there are various possible interpretations for the geometric data and / or road elements presented in the map fragment 400. For example, the first semantic map fragment 410 indicates the presence of three driving lanes, with one driving lane in the first direction 418A and two driving lanes in the second direction 418B. At the same time, the second semantic map fragment 420 indicates the presence of two lanes that merge into a single lane in the first direction 418A and two lanes that merge into a single lane in the second direction 418B. These different interpretations convey or illustrate different driving practices, rules, and habits, because three driving lanes are different from two merged segments. In the absence of such information, the map fragment 400 itself does not provide sufficient information for an entity to plan and utilize the road elements according to driving rules and habits. In contrast, each of the first semantic map fragment 410 and the second semantic map fragment 420 provides insights that enable an entity to plan and utilize the road elements according to driving rules and habits.

[0052] Figures 5A to 5C illustrates a non-limiting example of various map fragments related to a geographical area. More specifically, Figure 5A illustrates a map fragment 500, which is generated by the SLAM module 130 and includes geometric data. In this case, the geometric data includes a first lane marking 502, a second lane marking 504, a third lane marking 506, and a fourth lane marking 508. Although this map fragment 500 indicates the presence of the first lane marking 502, the second lane marking 504, the third lane marking 506, and the fourth lane marking 508, the map fragment 500 does not provide any meaningful information about the first lane marking 502, the second lane marking 504, the third lane marking 506, and the fourth lane marking 508.

[0053] Figure 5B illustrates a first semantic map fragment 510 associated with a geographical area that is the same as the geographical area of the map fragment 500 followed by Figure 5A In an example embodiment, the first semantic map fragment 510 is generated by the semantic map generator 120. In this regard, the system 100 is configured to generate the first semantic map fragment 510, for example, after verifying the estimate of the semantic road model based on traffic data, which indicates that in the corresponding part of the geographical area, there are typically two traffic flows that merge into a single lane with a similar priority. In addition, as Figure 5BAs shown, the first semantic map segment 510 provides meaningful information related to geometric data and / or road elements. In this regard, for example, the first semantic map segment 510 includes a first lane identifier 512 to identify the space between the first lane marking 502 and the second lane marking 504 as a driving lane. Additionally, the first semantic map segment 510 includes a second lane identifier 514 to identify the space between the third lane marking 506 and the fourth lane marking 508 as a driving lane. Additionally, the first semantic map segment 510 includes a third lane identifier 516 to identify the space between the first lane marking 502 and the fourth lane marking 508 as a driving lane. Further, the first semantic map segment 510 also indicates that the first driving lane 512 and the second driving lane 514 have equal priority with respect to merging into the third driving lane 516.

[0054] Figure 5C Illustrates a second semantic map segment 520 associated with a geographic region that is the same as the geographic region of the map segment 500 with which it is Figure 5A associated. In an example embodiment, the second semantic map segment 520 is generated by the semantic map generator 120. In this regard, the system 100 is configured to generate the second semantic map segment 520, for example, after validating an estimate of the semantic road model based on traffic data, which indicates that in the corresponding portion of the geographic region, there are typically two traffic flows that merge into a single lane, where the first lane appears to have priority over the second lane. Additionally, as Figure 5C shown, the second semantic map segment 520 provides meaningful information related to geometric data and / or road elements. Similar to the first semantic map segment 510, the second semantic map segment 520 includes a first lane identifier 512, a second lane identifier 514, and a third lane identifier 516. However, in contrast to the first semantic map segment 510, the second semantic map segment 520 indicates that when merging into the third lane 516, the first driving lane associated with the first lane identifier 512 has priority over the second driving lane associated with the second lane identifier 514. In this regard, compared to the merging of the first semantic map segment 510, the second semantic map segment 520 assigns a different meaning to the merging of the first driving lane and the second driving lane. Additionally, as Figure 5C shown, the second semantic map segment 520 identifies the dominant or primary driving direction of each of the first, second, and third driving lanes.

[0055] As Figure 5B and 5CAs shown, there are various possible interpretations for the geometric data and / or road elements presented in the map fragment 500. For example, the first semantic map fragment 510 indicates the presence of two lanes that merge into a single lane with equal priority. At the same time, the second semantic map fragment 520 indicates the presence of two lanes that merge into a single lane, where the first lane has priority over the second lane. These different interpretations convey or elaborate on different driving practices, rules, and habits. In the absence of such information, the map fragment 500 itself does not provide sufficient information for an entity to plan and utilize the road elements according to driving rules and habits. In contrast, each of the first semantic map fragment 510 and the second semantic map fragment 520 provides insights that enable an entity to plan and utilize the road elements according to driving rules and habits.

[0056] Figures 6A to 6C Illustrates a non-limiting example of various map fragments related to a geographic area. More specifically, Figure 6A Illustrates a map fragment 600 that is generated by the SLAM module 130 and includes geometric data. In this case, the geometric data includes a first lane marking 602, a second lane marking 604, a third lane marking 606, and a fourth lane marking 608. Additionally, the map fragment 600 indicates the presence of a left turn arrow 610 between the first lane marking 602 and the second lane marking 604. The map fragment 600 also indicates the presence of a first traffic light 612 and a second traffic light 614. Although this map fragment 600 indicates the presence of the first lane marking 602, the second lane marking 604, the third lane marking 606, the fourth lane marking 608, the left turn arrow 610, the first traffic light 612, and the second traffic light 614, the map fragment 600 does not provide any meaningful information about the geometric data and / or road elements.

[0057] Figure 6B Illustrates a first semantic map fragment 620 associated with the same geographic area as the geographic area of the map fragment 600 that Figure 6A In an example embodiment, the first semantic map fragment 620 is generated by the semantic map generator 120. In this regard, the system 100 is configured to generate, for example, the first semantic map fragment 620 after validating the estimates of the semantic road model based on traffic data, which indicates that in the corresponding part of the geographic area, the lanes turn left according to the traffic light, and two adjacent lanes exhibit similar behavior according to another traffic light. Additionally, as Figure 6BAs shown, the first semantic image segment 620 provides meaningful information related to geometric data and / or road elements. In this regard, for example, the first semantic image segment 620 includes a first lane identifier 622 to identify the space between the first lane marking 602 and the second lane marking 604 as a turning lane. Additionally, the first semantic image segment 620 includes a second lane identifier 624 to identify the space between the second lane marking 604 and the third lane marking 606 as a driving lane. Additionally, the first semantic image segment 620 includes a third lane identifier 626 to identify the space between the third lane marking 606 and the fourth lane marking 608 as a driving lane. Furthermore, the first semantic image segment 620 also indicates the relationship between the first traffic light 612 and the second traffic light 614 with respect to the first lane 622, the second lane 624, and the third lane 626. More specifically, as Figure 6B shown, the first semantic image segment 620 associates the first traffic light 612 with the first lane 622. Additionally, as Figure 6B shown, the first semantic image segment 620 associates the second traffic light 614 with the second lane 624 and the third lane 626. Additionally, the first semantic image segment 620 includes a driving direction for each of the first, second, and third lanes.

[0058] Figure 6C illustrates a second semantic image segment 640 associated with a geographic area that is the same as the geographic area of the map segment 600 that Figure 6A follows. In an example embodiment, the second semantic image segment 640 is generated by the semantic map generator 120. In this regard, the system 100 is configured to generate the second semantic image segment 640, for example, after validating the estimate of the semantic road model based on traffic data, which indicates that in the corresponding part of the geographic area, the lanes turn left according to the traffic lights, and two adjacent lanes exhibit different behaviors. Additionally, as Figure 6C shown, the second semantic image segment 640 provides meaningful information related to geometric data and / or road elements. Similar to the first semantic image segment 620, the second semantic image segment 640 includes a first lane identifier 622, a second lane identifier 624, and a third lane identifier 626. However, in contrast to the first semantic image segment 620, the second semantic image segment 640 provides a different relationship between the first traffic light 612 and the second traffic light 614 with respect to the first lane 622, the second lane 624, and the third lane 626. More specifically, as Figure 6C shown, the second semantic image segment 640 associates the first traffic light 612 only with the second lane 624. Additionally, as Figure 6CAs shown, the second semantic map segment 640 associates the second traffic light 614 only with the third lane 626. In addition, the second semantic map segment 640 introduces a third traffic light 642 that was not detected by the sensor during the data collection process 10 and is not provided as part of the geometric data of the map segment 600. Additionally, as Figure 6C shown, the second semantic map segment 640 associates the third traffic light 642 with the first lane 622. In this regard, the system 100 is configured to infer the existence of the third traffic light 642 based on an analysis of relevant data (e.g., traffic data, geometric data, etc.) for the geographic area. As demonstrated by this non-limiting example, the system 100 is configured to generate a semantic map and include undetected elements, such as the third traffic light 642, that may have been missed by one or more of the sensors due to obstacles, sensor errors, or various factors.

[0059] As Figure 6B and 6C shown, there are various possible interpretations for the geometric data and / or road elements presented in the map segment 600. For example, the first semantic map segment 620 indicates that the middle lane and the left lane are controlled by the same traffic light 614. At the same time, the second semantic map segment 640 indicates that each of the three lanes is controlled by its own traffic light. These different interpretations convey or elaborate on different driving practices, rules, and habits. In the absence of such information, the map segment 600 itself does not provide sufficient information for an entity to plan and utilize the road elements according to driving rules and habits. In contrast, each of the first semantic map segment 620 and the second semantic map segment 640 provides insights that enable an entity to plan and utilize the road elements according to driving rules and habits.

[0060] As described above, the system 100 provides many advantageous features and benefits. For example, the system 100 is advantageously configured to automatically or semi-automatically generate at least one estimate of a semantic road model. Additionally, the system 100 is configured to check the validity of the estimate of the semantic road model. Additionally, the system 100 is configured to provide statistical information regarding the correlations, confidences, and / or relationships between various map elements. Furthermore, the system 100 can be applied and / or implemented with a composite map model.

[0061] Additionally, system 100 is configured to generate a semantic map, which is advantageous in providing meaning and insights about various map elements. Additionally, system 100 is configured to provide an appropriate confidence level for semantic interpretation. Additionally, system 100 is configured to identify semantic elements that have relatively high uncertainty (in terms of the probability of existence of elements and relationships) or high ambiguity (in terms of element type or relationship classification), and request further data collection for these semantic elements. System 100 is also capable of providing automatic relevance detection, for example, by identifying semantic features that have little association with the observed traffic behavior as potentially unimportant. Additionally, system 100 is configured to quantify the data requirements for implementing an effective semantic road model by identifying relevant road segments with insufficient information while also requesting and / or obtaining the necessary data for implementing the effective semantic road model. Further, system 100 is configured to provide such automatic semantic map generation without manual processing.

[0062] That is, the foregoing description is intended to be illustrative rather than restrictive, and it is provided in the context of a particular application and its requirements. From the foregoing description, those skilled in the art can appreciate that the present invention can be implemented in various forms, and various embodiments can be implemented individually or in combination. Thus, although embodiments of the present invention have been described in conjunction with specific examples thereof, the general principles defined herein can be applied to other embodiments and applications without departing from the spirit and scope of the described embodiments, and the true scope of the embodiments and / or methods of the present invention is not limited to the embodiments shown and described, as various modifications will become apparent to those skilled in the art after studying the drawings, the specification, and the appended claims. For example, components and functionality can be separated or combined in ways different from those of the various embodiments described, and different terminology can be used to describe them. These and other variations, modifications, additions, and improvements can fall within the scope of the present disclosure as defined in the appended claims.

Claims

1. A method for automatic semantic map generation, comprising: generating a probabilistic road model via a processing system having at least one processing device based on road elements associated with geometric data corresponding to a geographic area; generating a probabilistic traffic model for the road model via the processing system; calculating a statistical inference result via the processing system based on the road model and the traffic model; and generating an estimate of a semantic road model via the processing system based on the statistical inference result, the method further comprising assuming a probabilistic road model for how road elements relate to each other and assuming a probabilistic traffic model for how traffic participants interact based on the probabilistic road model.

2. The method according to claim 1, wherein The statistical inference result is calculated via Bayesian inference or statistical hypothesis testing.

3. The method according to claim 1, wherein: the probabilistic road model includes a probabilistic structure having (i) identification data for identifying road elements based on geographic elements and (ii) relationship data for identifying relationships between the road elements; and the probabilistic traffic model includes a likelihood of traffic behavior for the road model.

4. The method according to claim 3, wherein: the identification data includes lane identifiers, shoulder identifiers, median identifiers, sign identifiers, and traffic light identifiers; the relationship data includes lane priority data, lane merge data, driving direction data, first association data, and second association data; the first association data includes an association between the sign identifier and the lane identifier; and the second association data includes an association between the traffic light identifier and the lane identifier.

5. The method according to claim 1, further comprising: generating a semantic map including the estimate of the semantic road model via the processing system.

6. The method according to claim 5, further comprising: storing the semantic map in at least one computer-readable medium accessible by a computer module operable to perform autonomous route planning based on the estimate of the semantic road model.

7. The method according to claim 1, further comprising: determining via the processing system whether the estimate of the semantic road model is invalid or valid; obtaining additional data related to the geographic area via the processing system when it is determined that the estimate of the semantic road model is invalid; and generating a semantic map including the estimate of the semantic road model via the processing system when it is determined that the estimate of the semantic road model is valid.

8. The method according to claim 7, wherein The step of determining whether the estimate of the semantic road model is invalid or valid includes evaluating a comparison result with a threshold criterion, the comparison being between traffic prediction data for the estimate of the semantic road model and traffic logs of actual traffic for at least one corresponding section of the geographic area.

9. The method according to claim 1, further comprising: obtaining geometric data for a geographic area via the processing system; and obtaining traffic data for the geographic area via the processing system; wherein: Generate the probabilistic road model based at least on the geometric data; and Generate the probabilistic traffic model based at least on the traffic data and the probabilistic road model.

10. A system for automatic semantic map generation, the system comprising: At least one computer-readable medium storing geometric data, traffic data, and computer-readable data; A processing system communicatively coupled to the at least one computer-readable medium, the processing system including at least one processing device and configured to execute the computer-readable data to implement a method including: Generate a probabilistic road model for road elements based on geometric data corresponding to a geographic region; Generate a probabilistic traffic model for the road model; Calculate a statistical inference result based on the road model and the traffic model; Generate an estimate of the semantic road model based on the statistical inference; And Generate a semantic map including the estimate of the semantic road model, Wherein the method further includes assuming a probabilistic road model for how road elements relate to each other and assuming a probabilistic traffic model for how traffic participants interact based on the probabilistic road model.

11. The system according to claim 10, wherein, The statistical inference result is calculated via Bayesian inference or a statistical hypothesis test.

12. The system according to claim 10, wherein: The probabilistic road model includes a probabilistic structure having (i) identification data identifying road elements based on the geometric data and (ii) relationship data indicating relationships between the road elements; and The probabilistic traffic model includes a likelihood of traffic behavior for the road model.

13. The system according to claim 10, wherein The processing system is configured to store the semantic map in the at least one computer-readable medium such that a computer module is operable to obtain the semantic map from the at least one computer-readable medium and perform autonomous route planning based on the semantic map.

14. A non-transitory computer-readable medium including computer-readable data, the computer-readable data being configured to implement a method including the following when executed by a processing system including at least one processing device: Generate a probabilistic road model for road elements based on geometric data corresponding to a geographic region; Generate a probabilistic traffic model for the road model; Calculate a statistical inference result based on the road model and the traffic model; And Generate an estimate of the semantic road model based on the statistical inference result, The method further includes assuming a probabilistic road model for how road elements relate to each other and assuming a probabilistic traffic model for how traffic participants interact based on the probabilistic road model.

15. The computer-readable medium according to claim 14, wherein The statistical inference result is calculated via Bayesian inference or a statistical hypothesis test.

16. The computer-readable medium according to claim 14, wherein: The probabilistic road model includes a probabilistic structure having (i) identification data identifying road elements based on the geometric data and (ii) relationship data indicating relationships between the road elements; and The probabilistic traffic model includes a likelihood of traffic behavior for the road model.

17. The computer-readable medium according to claim 14, wherein the method further comprises: generating, via the processing system, a semantic map including an estimate of the semantic road model.

18. The computer-readable medium according to claim 14, wherein the method further comprises: determining, via the processing system, whether the estimate of the semantic road model is invalid or valid; when it is determined that the estimate of the semantic road model is invalid, requesting, via the processing system, additional data related to the geographical area; and when it is determined that the estimate of the semantic road model is valid, generating, via the processing system, a semantic map including the estimate of the semantic road model.

19. The computer-readable medium according to claim 18, wherein, The step of determining whether the estimate of the semantic road model is invalid or valid includes evaluating a comparison result with a threshold criterion, the comparison being between traffic prediction data for the estimate of the semantic road model and traffic logs of actual traffic for at least one corresponding section of the geographical area.

20. The computer-readable medium according to claim 14, wherein the method further comprises: generating additional road elements based on traffic data to account for traffic behavior; wherein: the traffic data includes traffic logs of actual traffic for at least one corresponding section of the geographical area.

Images