Method and system for identifying static objects on a path

Abstract

Disclosed herein is a method and system for identifying one or more static objects on a path. In one embodiment, the method comprises generating a fused bin feature matrix corresponding to one or more static objects and determining an attention mask corresponding to the fused bin feature matrix. Further, the method comprises generating a transformed bin feature matrix by applying a series of multi-head attention blocks to the fused bin feature matrix and the attention mask. Subsequently, the edge feature matrix corresponding to the fused bin feature matrix is ​​determined by identifying one or more edge features corresponding to the fused bin feature matrix.The method further includes generating segment feature data by concatenating the transformed bin feature matrix with the edge feature matrix, and determining the output probability value for segment feature data by analyzing the segment feature data using a classification model. The navigation system further identifies one or more static objects along the path based on the determined output probability value.

Description

The following description describes in particular the invention and the manner in which it is intended to be practiced.The present disclosure relates generally to autonomous vehicle navigation. More particularly, the present disclosure relates to a method and system for identifying one or more static objects on a path of a navigating vehicle.Currently, all vehicles, in particular autonomous vehicles, are equipped with a plurality of sensors for detecting static obstacles on the path traveled by the vehicles. However, the static obstacles detected and interpreted by one sensor may deviate from the result of another sensor. For example, a lidar sensor detects the distance to a static obstacle in front of the vehicle better than a camera sensor. The camera sensor likewise recognizes the type of static obstacle in front of the vehicle much better than the lidar sensor. Each sensor detects the static obstacle in front of the vehicle and generates bin data including a set of bins corresponding to the static obstacle. Thus, the accuracy with which a static obstacle is detected largely depends on the accuracy of the fused bin data generated by combining bins of multiple sensors.Segmentation is a process in which a polygon plot of a static obstacle is generated by connecting data points of fused data from multiple sensors together. However, current segmentation methods sequentially generate polygon trains representing static obstacles by processing each fused data point using rule-based logic. The rule-based logic uses various manually fine-tuned parameters. For example, the rule-based logic may include a threshold distance between two bins of the fused bin data. If the threshold criteria are not met, the segmentation is not performed for the two bins of the fused bin data. This leads to irrevocable detections of static obstacles, which in turn can lead to accidents.Various techniques for identifying static objects are conventionally known. For example, Patent Publication No. CN113112093 discloses a method and an apparatus for detecting an abnormal object. The method includes capturing attribute information about an object in a target application at a plurality of different viewpoints of the scene. Further, the method includes extracting attention features of the associated object and the associated attribute information of the object at each viewpoint on the scene to obtain the attention feature information of the object at each viewpoint on the scene. In addition, the method includes performing information fusion processing on the attention feature information of the object at each viewpoint of the scene to obtain target feature information of the object. Finally, the method includes performing abnormality detection on the object according to the target feature information to obtain and output an abnormality detection result of the object.Another patent publication number CN110084299B discloses a method and apparatus for target recognition based on multi-head attention fusion. The method includes performing scale standardization on three feature maps having different scales of an image to be recognized to obtain three feature maps having the same scale. Further, the method includes performing multi-head attention fusion on the three feature maps at the same scale to obtain original prediction information. Still another patent publication number CN114663670A discloses a method for image recognition. The method includes determining a first vector sequence corresponding to an image block sequence corresponding to an image to be processed. Further, the method includes sequentially performing attention encoding on the first vector sequence based on a plurality of attention encoding layers to obtain a first feature map. In addition, the method includes determining an attention matrix corresponding to each attention encoding layer, extracting target attention features corresponding to the classified embedded vectors in each attention matrix, and performing fusion processing on each target attention feature to obtain a first attention distribution map. Moreover, the method includes performing negating processing on the first attention distribution map to obtain a reversed attention distribution map, and performing fusion processing on the reversed attention distribution map and the first feature map to obtain a second feature map. Finally, the method includes recognizing a target object corresponding to the first feature map, the first attention distribution map, and the second feature map to obtain the category information and the position information of the target object in the image to be processed.However, none of the conventional approaches discloses identifying static objects based on dynamic rule-based logic. Therefore, there is a need for a method and system for accurately identifying one or more static objects on the path of the navigating vehicle without using rule-based logic.The information disclosed in this Background of the Disclosure section is merely for enhancement of understanding of the general background of the invention and should not be taken as an acknowledgement or any form of suggestion that this information represents the prior art already known to a person skilled in the art.Disclosed herein is a method for identifying one or more static objects in a path. The method includes generating, by a navigation system, a fused bin feature matrix corresponding to one or more static objects, wherein the fused bin feature matrix includes a matrix of a plurality of bins and features of the plurality of bins. Further, the method includes determining, by the navigation system, an attention mask corresponding to the fused bin feature array by identifying one or more edge features corresponding to the fused bin feature array. The method also includes generating, by the navigation system, a transformed bin feature matrix by performing a multi-head attention analysis on the fused bin feature matrix and the attention mask. The method also includes determining, by the navigation system, an edge feature matrix corresponding to the fused bin feature matrix by identifying one or more edge features corresponding to the fused bin feature matrix. The method also includes generating, by the navigation system, segment feature data by concatenaten the transformed bin feature matrix with the edge feature matrix. Subsequently, the method includes determining, by the navigation system, an output probability value for the segment feature data by analyzing the segment feature data using a classification model. Finally, the method includes identifying, by the navigation system, one or more static objects on the path based on the determined output probability value.Further, the present disclosure relates to a navigation system for identifying one or more static objects on a path. The navigation system includes a processor and a memory. The memory is communicatively coupled to the processor. The memory stores processor-executable instructions that, when executed, cause the processor to generate a fused bin feature matrix corresponding to one or more static objects, the fused bin feature matrix comprising a matrix of a plurality of bins and features of the plurality of bins. Further, the processor determines an attention mask corresponding to the fused bin feature array by identifying one or more edge features corresponding to the fused bin feature array. In addition, the processor generates a transformed bin feature matrix by performing multi-head attention analysis on the fused bin feature matrix and the attention mask. The processor also determines an edge feature matrix corresponding to the fused bin feature matrix by identifying one or more edge features corresponding to the fused bin feature matrix. The processor also generates segment feature data by concatenaten the transformed bin feature matrix with the edge feature matrix. The processor then determines an output probability value for the segment feature data by analyzing the segment feature data using a classification model. Finally, the processor identifies one or more static objects on the path based on the determined output probability value.The foregoing summary is illustrative only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments and features described above, other aspects, embodiments and features will become apparent by reference to the drawings and the following detailed description.The accompanying drawings, which are incorporated in and constitute a part of this disclosure, illustrate exemplary embodiments and together with the description explain the principles disclosed. In the figures, the left(s) digit(s) of a reference sign indicate (indicate) the figure in which the reference sign appears for the first time. The same numerals are used throughout the figures to refer to similar features and components. Some embodiments of a system and / or method according to embodiments of the present subject matter will now be described, by way of example only, and with reference to the accompanying figures, in which: FIG. 1 illustrates an overview of an example environment in identifying one or more static objects in a path, according to embodiments of the present disclosure. FIG. 2 shows a detailed block diagram of a navigation system according to embodiments of the present disclosure. FIG. 3 illustrates a method for determining an output probability value, in accordance with embodiments of the present disclosure. FIG. 4 illustrates a flowchart illustrating a method for identifying one or more static objects in a path according to embodiments of the present disclosure. FIG. 5 shows a flow diagram illustrating a method for generating a fused bin feature matrix according to embodiments of the present disclosure. FIG. 6 illustrates a block diagram of an example computer system for implementing embodiments consistent with the present disclosure.It should be appreciated by those skilled in the art that all block diagrams included herein represent conceptual views of exemplary systems embodying the principles of the present subject matter. It should also be noted that all flowcharts, flowcharts, state transition diagrams, pseudo code, and the like represent various processes that may be substantially represented on a computer readable medium and executed by a computer or processor, regardless of whether that computer or processor is explicitly shown.In the present disclosure, the word "exemplary" is used to mean "serving as an example, case, or illustration.". An embodiment or implementation of the present subject matter described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments.In other words, the present disclosure proposes a method and system for generating a fused bin feature matrix corresponding to one or more static objects, wherein the fused bin feature matrix comprises a matrix of a plurality of bins and features of the plurality of bins. Further, the proposed method and system include determining an attention mask corresponding to the fused bin feature matrix by identifying one or more edge features corresponding to the fused bin feature matrix. By applying a series of multi-head attention blocks to the fused bin feature array and the attention mask, a transformed bin feature array is generated. In addition, the proposed method and system includes determining an edge feature matrix corresponding to the fused bin feature matrix by identifying one or more edge features corresponding to the fused bin feature matrix. Subsequently, the proposed method and system comprises generating segment feature data by concatenaten the transformed bin feature matrix with the edge feature matrix. The system determines an output probability value for the segment feature data by analyzing the segment feature data using a classification model, and finally identifies one or more static objects on the path based on the determined output probability value.In the following detailed description of the embodiments of the disclosure, reference is made to the accompanying drawings, which form a part hereof, and in which is shown by way of illustration specific embodiments in which the disclosure may be practiced. These embodiments are described in sufficient detail to enable those skilled in the art to practice the disclosure, and it is to be understood that other embodiments may be utilized and changes may be made without departing from the scope of the present disclosure. The following description is therefore not to be taken as limiting.FIG. 1 shows an example overview of an example environment 100 in identifying one or more static objects on a path 103 according to embodiments of the present disclosure.In one embodiment, as shown in the example environment 100, a vehicle 102 may move on a path 103. The vehicle 102 may be communicatively coupled to a navigation system 101. The navigation system 101 may be any computing device configurable to identify one or more static objects, namely, the static object 1 104 a, the static object 2 104 b, and the static object 3 104 c(collectively referred to as static objects 104) present on the path 103. The one or more static objects may include, but are not limited to, a bridge, handrail, sidewalk, barrier, etc. The vehicle 102 may be an autonomous vehicle, such as, but is not limited to, a 'Level 0 vehicle - no automated driving', a 'Level 1 vehicle - driver assistance', a 'Level 2 vehicle - partially automated driving', a 'Level 3 vehicle - conditionally automated driving', a 'Level 4 vehicle - highly automated driving', or a 'Level 5 vehicle - fully automated driving'. The vehicle 102 may be equipped with a plurality of sensors that provide real-time input of sensor data pertaining to the path 103 and one or more vehicles, objects, or obstacles on the path 103.In one implementation, the navigation system 101 may be implemented external to the vehicle 102 (e.g., on a cloud server) and connected to the vehicle 102 via a predefined wireless communication network. In an alternative implementation, the navigation system 101 may be deployed in the vehicle 102 as part of an existing electronic control unit (ECU) of the vehicle 102 and / or as a stand-alone ECU of the vehicle 102.In one embodiment, the plurality of sensors equipped to the vehicle 102 detect the one or more static objects on the path 103 of the navigating vehicle 102. Further, the navigation system 101 may receive the static object related data from the plurality of sensors and generate a fused bin feature matrix corresponding to one or more static objects, wherein the fused bin feature matrix comprises a matrix of a plurality of bins and features of the plurality of bins. In an embodiment, the features of the plurality of bins may include, but are not limited to, coordinate data of the bins, a distance of an origin from the bins, and sequential probability ratio test (SRT) class information, and an angle spanned to the lateral axis of the vehicle. In one embodiment, the origin may represent a center of the navigating vehicle 102.After generating the fused bin feature array, the navigation system 101 may determine an attention mask corresponding to the fused bin feature array by identifying one or more edge features corresponding to the fused bin feature array. In one embodiment, the attention mask may include, but is not limited to, a distance between two bins, an angle spanned at the origin, and sensor connection data. Further, in one embodiment, the attention mask encodes known information about sensor-determined connections between bins. In this case, the bin-bin connections identified by sensors are assigned a higher weighting, and bin-bin connections in which the bins originate from different SRT classes are assigned a negative weighting.In one embodiment, after the navigation system 101 has determined the attention mask, it may generate a transformed bin feature matrix by applying a series of multi-head attention blocks to the fused bin feature matrix and the attention mask.After generating the transformed bin feature matrix, the navigation system 101 may determine an edge feature matrix corresponding to the fused bin feature matrix by identifying one or more edge features corresponding to the fused bin feature matrix. In one embodiment, the edge feature matrix may include, but is not limited to, a distance between two bins, an angle spanned at the origin, and sensor connection data.In one embodiment, after the navigation system 101 determines the edge feature matrix, it may generate segment feature data by concatenates the transformed bin feature matrix with the edge feature matrix.After obtaining the segment feature data, the navigation system 101 may obtain an output probability value for the segment feature data by analyzing the segment feature data using a classification model. In an embodiment, the output probability value corresponds to a probability of forming a segment between two bins among the plurality of bins of the transformed bin feature matrix.In one embodiment, after the navigation system 101 determines the output probability value, it may identify one or more static objects on the path 103 based on the determined output probability value.Finally, based on the foregoing analysis of the bins, the navigation system 101 may accurately identify the one or more static objects on the path 103 of the navigating vehicle 102. After identifying one or more static objects on the path 103, the vehicle 102 may change its navigation direction to avoid accidents, for example. The navigation system 101 thus helps identify one or more static objects by fusing the properties detected using multiple sensors.FIG. 2 shows a detailed block diagram of the navigation system 101 in accordance with embodiments of the present disclosure.In some implementations, the navigation system 101 may include an I / O interface 201, a processor 203, and a memory 205. The I / O interface 201 may be communicatively coupled to the vehicle 102 via an interface to gather various information related to navigation of the vehicle 102. The memory 205 may be communicatively coupled to the processor 203 and the one or more modules 209, and store data 207. The processor 203 may be configured to perform one or more functions of the navigation system 101 in identifying one or more static objects on the path 103 using the data 207 and one or more modules 209.In one embodiment, data 207 stored in memory 205 may include, but is not limited to, fused bin feature matrix 211, attention mask 213, transformed bin feature matrix 215, edge feature matrix 217, segment feature data 219, output probability value 221, and remaining data 223. In some implementations, data 207 may be stored in memory 205 in the form of various data structures. Additionally, the data 207 may be organized using data models, such as relational or hierarchical data models. The remaining data 223 may include the intermediate values and temporary data generated when identifying one or more static objects on the paths 103.The one or more modules 209, in one embodiment, may include, but are not limited to, a fused bin feature matrix generation module 231, an attention mask determination module 233, a transformed bin feature matrix generation module 235, an edge feature matrix determination module 237, a segment feature data generation module 239, a probability score determination module 241, a static object identification module 243, and other modules 245. The other modules 245 may include any component or module used to identify one or more static objects on the paths 103.In one embodiment, fused bin feature matrix generation module 231 may be configured to generate fused bin feature matrix 211 corresponding to the one or more static objects. Further, in one embodiment, fused bin feature matrix 211 comprises a matrix of a plurality of bins and features of the plurality of bins. The fused bin feature matrix 211 matrix may include, as an example, a plurality of bins in rows and features of a plurality of bins in corresponding columns. In an embodiment, the features of the plurality of bins may include, but are not limited to, X coordinate data of the bins, Y coordinate data of the bins, a distance of the bin from the origin, sequential probability ratio test (SRT) information on the bin, shape type information of the bin, and the angle spanned to the lateral axis of the vehicle.In one embodiment, the attention mask determination module 233 determines the attention mask 213 corresponding to the fused bin feature array 211 by identifying one or more edge features corresponding to the fused bin feature array 211. In an embodiment, attention mask 213 may include, but is not limited to, a distance between two bins, an angle spanned at the origin, a segment type, and sensor connection data. In one embodiment, the edge features corresponding to fused bins feature matrix 211 are obtained from the sensor data of the plurality of sensors. Further, in one embodiment, attention mask 213 encodes known information about sensor-determined connections between bins. In this case, the bin-bin connections identified by sensors are assigned a higher weighting, and bin-bin connections in which the bins originate from different SRT classes are assigned a negative weighting.In one embodiment, the transformed bin feature matrix generation module 235 may be configured to generate a transformed bin feature matrix 215 by applying a series of multi-head attention blocks to the fused bin feature matrix 211 and the attention mask 213. In one embodiment, fused bin feature matrix 211 comprising the features of the plurality of bins and attention mask 213 are embedded in the block of multi-head attention analysis to generate transformed bin feature matrix 215. The transformed bin feature matrix 215 provides the better representation of the bins of the fused bin feature matrix 211, and is also helpful in extracting better features of the plurality of bins of the fused bin feature matrix 211.In one embodiment, the edge feature array determination module 237 is configured to determine an edge feature array 217 corresponding to the fused bin feature array 211 by identifying one or more edge features corresponding to the fused bin feature array 211. In an embodiment, the edge feature matrix 217 may include, but is not limited to, a distance between two bins, an angle spanned at the origin, and sensor connection data.In one embodiment, the segment feature data generation module 239 generates the segment feature data 219 by concatenaten the transformed bin feature matrix 215 with the edge feature matrix 217. In one embodiment, the updated bin features in the transformed bin feature matrix 215 may be represented as "h". Further, the new bin representation of the ith bin may be indicated as h i. Further, the new bin representations of different bins of the transformed bin feature matrix 215 may be split into different SRT classes. This is because segments may be present between bins belonging to similar SRT classes.As an example, assume that SRT class A has 3 bins, such as h i, h j, h k, so that all 3 bins belong to similar SRT classes. To form a segment, the navigation system 101 must take and concatenate pairwise features of the bins such as h i, h j and the edge feature array represented as e ij to generate the segment feature data 219. The segment feature data may be represented as follows:In one embodiment, the probability value determination module 241 may be configured to determine an output probability value 221 for the segment feature data 219 by analyzing the segment feature data 219 using a classification model. The classification model may include, but is not limited to, a random forest classification model. In one embodiment, the output probability value 221 corresponds to the probability of forming a segment between two bins among the plurality of bins of the transformed bin feature matrix 215.In one embodiment, the static object identification module 243 identifies one or more static objects on the path 103 based on the determined output probability value 221.FIG. 3 is an exemplary diagram illustrating a method for determining an output probability value 221 according to embodiments of the present disclosure.As illustrated in FIG. 3, the fused bin feature array 211 and attention mask 213 are placed in a block of multiple multi-head attention analyses 301. In one embodiment, fused bin feature matrix 211 comprises a matrix of a plurality of bins and features of the plurality of bins. Further, in one embodiment, attention mask 213 may include, but is not limited to, a distance between two bins, an angle spanned at the origin, a segment type, and sensor connection data. The attention mask 213 encodes known information about connections between bins determined by the sensor. In this case, the bin-bin connections identified by sensors are assigned a higher weighting, and bin-bin connections in which the bins originate from different SRT classes are assigned a negative weighting. The multi-head attention analysis 301 is performed on the fused bin feature array 211 and the attention mask 213 to generate the transformed bin feature array 215. Additionally, in one embodiment, the multiple multi-head attention analysis 301 may include, but is not limited to, a 1st multi-head attention analysis 303, a 2nd multi-head attention analysis 305, a 3rd multi-head attention analysis 307, etc., as shown in FIG. 3.For example, consider the 1st multi-head attention analysis 303: the features of the plurality of bins of fused bins feature matrix 211 are embedded in the linear layers of the 1st multi-head attention analysis 303 block in a vectorized format, represented as X1, where the linear layers include layers of a query vector (Q), a value vector (V), and a key vector (K). The output of the linear layers of the query vector (Q), the value vector (V) and the key vector (K) of the block of the 1st multi-head attention analysis 303 can be referred to as QW Q1, VW V1 and VW, respectively. KW K1 can be used. Furthermore, for each output of the linear layers, attention analysis is performed by means of a scaled dot product. Therefore, the output of the 1st multi-head attention analysis can be represented as follows:In one embodiment, after the determination of the output of the block of the 1st multi-head attention analysis 303, the 2nd multi-head attention analysis 305 is carried out. In the block of the 2nd Multi Attention Analysis 305, the query vector (Q) is represented as X1, while the value vector (V) and key vector (K) are represented as X1+Y1. The output of the linear layers of the query vector (Q), the value vector (V) and the key vector (K) of the block of the 2nd multi-head attention analysis 305 may be referred to as QW Q2, VW V2 and VW, respectively. KW K2 can be used. Furthermore, for each output of the linear layers, attention analysis is performed by means of a scaled dot product. Therefore, the output of the 2nd multi-head attention analysis can be represented as follows:Accordingly, the output of the block of the N-th multi-head attention analysis may be represented as follows:After generating the transformed bin feature matrix 215, the navigation system 101 concatenates the transformed bin feature matrix 215 with the edge feature matrix 217 to generate the segment feature data 219. In an embodiment, the edge feature matrix 217 may include, but is not limited to, a distance between two bins, an angle spanned at the origin, and sensor connection data.Further, in one embodiment, the segment feature data 219 is passed to a classification model 309 and the navigation system 101 determines the output probability value 221 for the segment feature data 219 by analyzing the segment feature data 219 using the classification model 309.FIG. 4 illustrates a flow diagram illustrating a method for identifying one or more static objects on a path 103 according to embodiments of the present disclosure.In an embodiment, as shown in FIG. 4, at block 401, the method 400 includes generating, by the navigation system 101, fused bin feature matrix 211 corresponding to the one or more static objects. Fused bin feature matrix 211 comprises a matrix of a plurality of bins and features of the plurality of bins. In one embodiment, the features of the plurality of bins may include, but are not limited to, coordinate data of the bins, a distance of an origin from the bins, and sequential probability ratio test (SRT) class information, and the angle spanned from the lateral axis of the vehicle. In one embodiment, the sensors of the vehicle 102 may detect the static object on the path 103. After detecting the static objects, the navigation system 101 may generate a fused bin feature matrix 211 comprising a plurality of bins. This is achieved by first detecting the detection of the static object across a plurality of sensors and then combining these detections. Further, the navigation system 101 may identify the features of the plurality of bins in the fused bin feature array 211. For example, the coordinate data of the bins may include, but are not limited to, an x-coordinate and a y-coordinate of the bins with respect to various openings of the grid. In addition, the origin of the bins may indicate a center of the vehicle 102. The distance between the origin and each of the plurality of bins may be estimated by the navigation system 101. Based on the above analysis, the navigation system 101 may generate the fused bin feature matrix 211 corresponding to the one or more static objects.At block 402, the method 400 includes determining, by the navigation system 101, the attention mask 213 corresponding to the fused bin feature array 211 by identifying one or more edge features corresponding to the fused bin feature array 211. In an embodiment, attention mask 213 may include, but is not limited to, a distance between two bins, an angle spanned at the origin, and sensor connection data. Sensor link data indicates whether or not the bins of fused bins feature array 211 belong to the same static object. In one embodiment, the edge features corresponding to fused bins feature matrix 211 may be obtained from the sensor data of the plurality of sensors. The navigation system 101 may determine the distance between two bins. Additionally, the navigation system 101 may identify the angle spanned by the origin of each of the plurality of bins of the fused bin feature matrix 211. Based on the above analysis, the navigation system 101 may determine the attention mask 213 corresponding to the fused bin feature matrix 211.At block 403, the method 400 includes generating, by the navigation system 101, the transformed bin feature matrix 215 by applying a series of multi-head attention blocks 301 to the fused bin feature matrix 211 and the attention mask 213.At block 404, the method 400 includes determining, by the navigation system 101, the edge feature matrix 217 corresponding to the fused bin feature matrix 211 by identifying one or more edge features corresponding to the fused bin feature matrix 211. In an embodiment, the edge feature matrix 217 may include, but is not limited to, a distance between two bins, an angle spanned at the origin, and sensor connection data. The navigation system 101 may determine the distance between two bins. Additionally, the navigation system 101 may identify the angle spanned by the origin of each of the plurality of bins of the transformed bin feature matrix 215. Based on the above analysis, the navigation system 101 may determine the edge feature matrix 217 corresponding to fused bin feature matrix 211.At block 405, the method includes generating, by the navigation system 101, the segment feature data 219 by concatenate the transformed bin feature matrix 215 with the edge feature matrix 217.At block 406, the method 400 includes determining, by the navigation system 101, an output probability value 221 for the segment feature data 219 by analyzing the segment feature data 219 using a classification model 309. In an embodiment, the output probability value 221 corresponds to the probability of forming a segment between two bins among the plurality of bins.At block 407, the method 400 includes identifying, by the navigation system 101, one or more static objects on the path 103 based on the determined output probability value 221.FIG. 5 shows a flow diagram illustrating a method 500 for generating a fused bin feature matrix 211, in accordance with embodiments of the present disclosure.At block 501, the method 500 includes generating, by the navigation system 101, a first set of bins corresponding to the one or more static objects using data collected from a first set of sensors of the vehicle 102. In one embodiment, the first set of sensors may include, but is not limited to, a camera sensor.At block 502, the method 500 includes generating, by the navigation system 101, a second set of bins corresponding to the one or more static objects using data collected from a second set of sensors of the vehicle 102. In one embodiment, the second set of sensors may include, but is not limited to, a lidar sensor and a radar sensor.At block 503, the method 500 includes generating, by the navigation system 101, the fused bin feature matrix 211 by fusing the first set of bins and the second set of bins. In one embodiment, fused bin feature matrix 211 comprises a matrix of a plurality of bins and features of the plurality of bins.Computer SystemFIG. 6 illustrates a block diagram of an example computer system 601 for implementing embodiments consistent with the present disclosure. In one embodiment, computer system 601 may be navigation system 101 illustrated in FIGS. 1 and 2, which is used to identify one or more static objects on a path 103 traveled by a vehicle 102. The computer system 601 may include a central processing unit ("CPU" or "processor") 602. Processor 602 may include at least one data processor for executing program components for executing user- or system-generated operating processes. The processor 602 may include specialized processing units such as integrated system (bus) controllers, memory management controllers, floating point units, graphics processing units, digital signal processing units, etc.Processor 602 may be in communication with one or more input / output (I / O) devices (604 and 605) via I / O interface 603. In some embodiments, processor 602 may be in communication with communication network 608 via network interface 607. The network interface 607 may communicate with the communication network 608. Using network interface 607 and communication network 608, computer system 601 may connect to vehicle 102 to acquire information regarding vehicle 102 and identify one or more static objects on path 103.In one implementation, the communication network 608 can be implemented as one of several types of networks, such as intranet or local area network (LAN) and the like within the organization. The communication network 608 can be either a dedicated network or a shared network that represents an association of multiple types of networks that use different protocols. In some embodiments, processor 602 may be in communication with memory 615 (e.g., RAM 613, ROM 614, etc., as shown in FIG. 6 ) via a memory interface 612.The memory 615 may store a collection of program or database components including, but not limited to, an operating system 616, a user / application interface 617, a web browser 618, a mail server 619, a mail client 620, and user / application data 621, and the like. In some embodiments, computer system 601 may store user / application data 621, such as the data, variables, records, etc., as described in this disclosure.Operating system 616 may enable resource management and operation of computer system 601. The user interface 617 may enable the display, execution, interaction, manipulation, or operation of program components through text or graphics functions. For example, user interface 617 may provide computer interaction interface elements on a display system operatively connected to computer system 601.The web browser 618 may be an application for displaying hypertext. Secure browsing on the Internet can be guaranteed through the use of secure hypertext transport protocol (HTTPS), secure sockets layer (SSL), transport layer security (TLS) and the like.In one embodiment, the method of the present disclosure helps accurately track static objects in front of the navigating vehicle by fusing the characteristics of data obtained from multiple sensors, thereby reducing the risk of an accident.In one embodiment, the method of the present disclosure enables better representation of the bins by applying a series of multi-head attention blocks to the bins.In addition, the method of the present disclosure helps improve the accuracy of identifying a static obstacle, and the proposed method is easily adaptable to previously unknown scenarios. Further, the method of the present disclosure helps in estimating the parameters obtained from data acquired from sensors and obviates manual fine tuning of the parameters. In addition, the method of the present disclosure facilitates parameter expansion if new sensors are added to the vehicle.As discussed above, it should be appreciated that the method and navigation system of the present disclosure may be used to address various technical issues associated with identifying one or more static objects in the path. That is, the above technical advances and practical applications of the proposed method may be attributed to aspects of generating segment feature data and determining the probability value for segmenting prediction of bins.In view of the technical advances achieved by the disclosed method and navigation system, the claimed steps, as discussed above, are not routine, conventional, or well known aspects of the prior art, as the claimed steps provide the above solutions to the technical problems existing in conventional technologies. Furthermore, the claimed steps clearly improve the operation of the system itself, as the claimed steps provide a technical solution to a technical problem.Although various aspects and embodiments have been disclosed herein, other aspects and embodiments will be apparent to those skilled in the art. The various aspects and embodiments disclosed herein are illustrative only and are not to be taken as limiting; the true scope and spirit is indicated by the following claims.List of reference numbers:100 Example environment 101 Navigation system 102 Vehicle 103 Path 104 aStatic object 1 104 bStatic object 2 104 cStatic object 3 201 I / O interface 203 Processor 205 Memory 207 Data 209 Module 211 Feature matrix of fused bins 213 Attention mask 215 Transformed bin feature matrix 217 Edge feature matrix 219 Segment feature data 221 Output probability value 223 Remaining data 231 Module for generating the feature matrix of fused bins 233 Module for ascertaining the attention mask 235 Module for generating the transformed bin feature matrix 237 Module for ascertaining the edge feature matrix 239 Module for generating segment feature data 241 Module for probability value ascertainment 243 Module for identifying static objects 245 Other modules 301 Multi-head attention analysis 303 1. Multi-Head Attention Analysis 305 2. Multi-Head Attention Analysis 307 3. Multi-Head Attention Analysis 309 Classification model 601 Exemplary Computer System 602 Processor of the exemplary Computer System 603 I / O interface of the exemplary Computer System 604 Input devices 605 Output devices 606 Transmitter / receiver of the exemplary Computer System 607 Network interface 608 Communication network 612 Memory interface 613 RAM 614 ROM 615 Memory of the exemplary Computer System 616 Operating system 617 User / application interface 618 Web browser 619 Mail server 620 Mail client 621 User / application dataReferences included in the specificationThis list of documents cited by the applicant has been produced in an automated manner and is only included for the better information of the reader. The list is not part of the German patent application or utility model application. The DPMA does not take any adhesion for any faults or omissions.Patent Literature citedCN 113112093

[0005] CN 110084299B

[0006] CN 114663670A

[0006]

Claims

A method for identifying one or more static objects on a path, the method comprising: generating, by a navigation system, a fused bin feature matrix corresponding to one or more static objects, wherein the fused bin feature matrix comprises a matrix of a plurality of bins and features of the plurality of bins; determining, by the navigation system, an attention mask corresponding to the fused bin feature matrix by identifying one or more edge features corresponding to the fused bin feature matrix; generating, by the navigation system, a transformed bin feature matrix by applying a series of multi-head attention blocks to the fused bin feature matrix and the attention mask; determining, by the navigation system, an edge feature matrix corresponding to the fused bin feature matrix by identifying one or more edge features corresponding to the fused bin feature matrix; generating, by the navigation system, segment feature data by concatenaten the transformed bin feature matrix with the edge feature matrix; determining, by the navigation system, an output probability value for the segment feature data by analyzing the segment feature data using a classification model; and identifying, by the navigation system, one or more static objects on the path based on the determined output probability value.The method of claim 1, wherein generating the fused bin feature array comprises: generating, by the navigation system, a first set of bins corresponding to the one or more static objects using data collected from a first set of sensors of the vehicle; generating, by the navigation system, a second set of bins corresponding to the one or more static objects using data collected from a second set of sensors of the vehicle; and generating, by the navigation system, the fused bin feature array by fusing the first set of bins and the second set of bins.The method of claim 2, wherein the first set of sensors comprises at least one camera sensor and the second set of sensors comprises at least one lidar sensor and a radar sensor.The method of claim 1, wherein the features of the plurality of bins comprise at least one of: coordinate data of the bins, distance of an origin from the bins, and sequential probability ratio test (SRT) class information.The method of claim 1, wherein the attention mask and the edge feature matrix comprise at least a distance between two bins, an angle spanned at the origin, and sensor connection data.The method of claim 1, wherein the output probability value corresponds to a probability that a segment is formed between two bins among the plurality of bins.A navigation system for identifying one or more static objects on a path, the navigation system comprising: a memory; and a processor communicatively coupled to the memory and configured to: generate a fused bin feature matrix corresponding to one or more static objects, the fused bin feature matrix comprising a matrix of a plurality of bins and features of the plurality of bins; determine an attention mask corresponding to the fused bin feature matrix by identifying one or more edge features corresponding to the fused bin feature matrix; generate a transformed bin feature matrix by applying a series of multi-head attention blocks to the fused bin feature matrix and the attention mask; determining an edge feature matrix corresponding to the fused bin feature matrix by identifying one or more edge features corresponding to the fused bin feature matrix; generating segment feature data by concatenaten the transformed bin feature matrix with the edge feature matrix; determining an output probability value for the segment feature data by analyzing the segment feature data using a classification model; and identifying one or more static objects on the path based on the determined output probability value.The navigation system of claim 7, wherein to generate the fused bin feature array, the processor is further configured to: generate a first set of bins corresponding to the one or more static objects using data collected from a first set of sensors of the vehicle; generate a second set of bins corresponding to the one or more static objects using data collected from a second set of sensors of the vehicle; and generate the fused bin feature array by fusing the first set of bins and the second set of bins.The navigation system of claim 8, wherein the first set of sensors comprises at least one camera sensor and the second set of sensors comprises at least one lidar sensor and a radar sensor.The navigation system of claim 7, wherein the features of the plurality of bins comprise at least one of: coordinate data of the bins, distance of an origin from the bins, and sequential probability ratio test (SRT) class information.The navigation system of claim 7, wherein the attention mask and the edge feature matrix comprise at least a distance between two bins, an angle spanned at the origin, and sensor connection data.The navigation system of claim 7, wherein the output probability value corresponds to a probability that a segment is formed between two bins among the plurality of bins.

Images