CN118243131B - Track prediction method based on vehicle track semantic analysis and deep belief network - Google Patents

Abstract

The invention relates to the field of track prediction, and discloses a track prediction method based on vehicle track semantic analysis and a deep belief network, which comprises the following steps: acquiring a historical driving track set; carrying out semantic analysis on the historical driving track set to obtain a resident point set; sorting according to the average residence time length and residence frequency corresponding to the residence points to obtain a residence point sequence, and selecting a first number of residence points with front sorting; generating prompt information based on the candidate resident point set and sending the prompt information to terminal equipment corresponding to a plurality of riding users; planning paths according to the target starting point, the target ending point, a plurality of target residence points and the pre-configured road network data to obtain a plurality of planned paths; acquiring a real-time traffic flow thermodynamic diagram and inputting the thermodynamic diagram into a deep belief network to obtain real-time road condition information; and selecting a target planning path from the plurality of planning paths according to the real-time road condition information and sending the target planning path to the terminal equipment. Therefore, track prediction in a school bus or a class bus pick-up scene is realized.

Description

Track prediction method based on vehicle track semantic analysis and deep belief network

Technical Field

The invention relates to the field of track prediction, in particular to a track prediction method based on vehicle track semantic analysis and a deep belief network.

Background

The wide application of various map software (APP) provides great convenience for the acquisition of vehicle track data. And the collection of a large amount of vehicle track data has important significance for track prediction.

The existing track prediction method often has the following technical problems:

First, the existing track prediction method is often a general track prediction method, and lacks a track prediction method suitable for special car scenes. Specifically, the school bus pick-up scene is a common use scene in practice. The specificity of the school bus pick-up scene is represented by the relatively fixed start and end points, the need for multiple stops during travel, relatively fixed time, etc. The universal track prediction method does not fully consider the particularities, so that the adaptation degree with a specific vehicle scene is not high, and the requirements of the specific vehicle scene cannot be met;

secondly, under the school bus pick-up scene, as the riding users are juveniles, how to ensure the safety of the riding users through technical means is a problem to be solved;

Thirdly, in the process of carrying out semantic analysis on the historical driving track set to obtain the stay points, the problem of stay point redundancy exists.

Disclosure of Invention

This summary is provided to introduce a selection of concepts in a simplified form that are further described below in the detailed description. This summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used to limit the scope of the claimed subject matter.

The invention proposes a track prediction method according to vehicle track semantic analysis and a deep belief network to solve one or more of the technical problems mentioned in the background section above.

The invention provides a track prediction method based on vehicle track semantic analysis and a deep belief network, which comprises the following steps: acquiring a historical driving track set of a target vehicle in a historical time period, wherein each historical driving track in the historical driving track set meets at least one of the following: the method comprises the steps of including a target starting point and a target ending point; carrying out semantic analysis on each historical driving track in the historical driving track set to obtain a plurality of resident points corresponding to each historical driving track, wherein the plurality of resident points corresponding to each historical driving track in the historical driving track set respectively form a resident point set; determining average residence time length and residence frequency corresponding to residence points in the residence point set, sorting according to the average residence time length and residence frequency to obtain a residence point sequence, selecting a first number of residence points with front sorting from the residence point sequence, and determining the first number of residence points as candidate residence point sets; generating prompt information based on the candidate resident point set, and sending the prompt information to terminal equipment respectively corresponding to a plurality of vehicle users, so that the plurality of vehicle users respectively select corresponding target resident points to obtain a plurality of target resident points and the number of vehicle users corresponding to each target resident point; planning paths according to the target starting point, the target ending point, a plurality of target residence points and the pre-configured road network data to obtain a plurality of planned paths; acquiring a real-time traffic flow thermodynamic diagram, and inputting the real-time traffic flow thermodynamic diagram into a pre-trained deep belief network to obtain real-time road condition information; selecting a target planning path from the plurality of planning paths according to the real-time road condition information; and sending the target planning path as a predicted track to terminal equipment respectively corresponding to a plurality of riding users.

Optionally, before sending the target planned path as the predicted track to terminal devices corresponding to the plurality of riding users respectively, the track prediction method further includes: generating prediction time for reaching each target residence point according to real-time road condition information, target planning paths and the number of passengers corresponding to each target residence point; and sending the target planning path as a predicted track to terminal equipment respectively corresponding to a plurality of riding users, wherein the terminal equipment comprises: and for each of the plurality of riding users, sending the predicted time and the target planning path corresponding to the target residence point selected by the riding user to the terminal equipment corresponding to the riding user.

Optionally, the terminal device is further configured to collect a face image of the riding user as a reserved face image when the riding user selects the target residence point; the track prediction method further comprises the following steps: when the target vehicle is detected to reach each target resident point, acquiring face images of each actual riding user of the target resident point through an image acquisition device arranged at an upper door of the target vehicle, so as to obtain a plurality of actual face images; comparing the actual face images with reserved face images of a plurality of riding users selecting target residence points to obtain comparison results; if the comparison result represents that a plurality of bus users of the target residence point are selected to have bus users which are not on bus, determining the bus users which are not on bus as target bus users, judging whether the target riding user is an underage user or not according to the user information of the target riding user, and if the target riding user is the underage user, sending prompt information representing that the target riding user is not on the bus to a terminal corresponding to the associated user of the target riding user.

Optionally, the track prediction method further includes: in the driving process of the target vehicle, acquiring images and voice data of a driver through an image acquisition device and a voice acquisition device which are arranged at a driving position of the target vehicle; respectively extracting features of the driver image and the driver voice data to obtain image features and voice features; and carrying out security assessment according to the image features and the voice features to obtain security assessment scores, and generating risk prompt information representing that the security risk exists if the security assessment scores are smaller than a preset score threshold value.

Optionally, each historical driving track comprises a track point sequence, track points in the track point sequence are triplets, and the triplets comprise longitude, latitude and time stamps; carrying out semantic analysis on each historical driving track in the historical driving track set to obtain a plurality of residence points corresponding to each historical driving track, wherein the semantic analysis comprises the following steps: for each historical driving track, determining whether a preset number of track points meeting a first preset condition exist in a track point sequence included in the historical driving track, wherein the first preset condition is that the preset number of track points are continuous track points, and the distance between any two adjacent track points in the preset number of track points is smaller than a preset distance threshold; if the historical driving track exists, determining the resident points corresponding to the preset number of track points, and obtaining a plurality of resident points corresponding to each historical driving track.

Optionally, determining an average residence time length and residence frequency corresponding to residence points in the residence point set, and sorting according to the average residence time length and residence frequency to obtain a residence point sequence, including: selecting one resident point from the resident point set, determining whether at least one resident point with the distance smaller than a first distance exists in the resident point set, if so, determining the selected resident point and the at least one resident point as a resident point group, determining the ratio between the number of the resident points in the resident point group and the number of the historical driving tracks in the historical driving track set, determining the ratio as the resident frequency, and calculating the average value of resident duration corresponding to each resident point in the resident point group to obtain the average resident duration; merging all resident points in the resident point group to obtain an updated resident point set; and sequencing all the resident points in the updated resident point set according to the average resident duration and resident frequency to obtain a resident point sequence.

The invention has the following beneficial effects: through carrying out semantic analysis on the historical driving track set, a plurality of corresponding resident points are extracted, and then candidate resident points are determined, so that more accurate resident point identification and automatic extraction are realized, and the requirements of a school bus pick-up scene are met. In the process, the average residence time and residence frequency of the residence points are fully considered, so that reasonable planning of the residence points is realized, the residence points are not required to be set manually, and a foundation is laid for the application of automatic driving in a school bus pick-up scene. In addition, in the process of generating the predicted track, the depth belief network is used for extracting real-time traffic thermodynamic diagrams to obtain real-time road condition information, so that high-order features in the image can be extracted, and the path selection is assisted by the real-time road condition information, so that the predicted track is generated. In the process, the deep belief network can fully extract the high-order characteristics of the real-time traffic thermodynamic diagram, so that more accurate real-time road condition information is generated, and the generated prediction track is more accurate.

Drawings

The above and other features, advantages and aspects of embodiments of the present invention will become more apparent by reference to the following detailed description when taken in conjunction with the accompanying drawings. The same or similar reference numbers will be used throughout the drawings to refer to the same or like elements. It should be understood that the figures are schematic and that elements and components are not necessarily drawn to scale.

FIG. 1 is an exemplary flow chart of a track prediction method according to the vehicle track semantic analysis and deep belief network of the present invention.

Detailed Description

The invention will be described in more detail below with reference to the accompanying drawings. While certain embodiments of the invention have been illustrated in the accompanying drawings, it is to be understood that the invention may be embodied in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete. It should be understood that the drawings and embodiments of the invention are for illustration purposes only and are not intended to limit the scope of the present invention.

It should be noted that, for convenience of description, only the portions related to the present invention are shown in the drawings. Embodiments of the invention and features of the embodiments may be combined with each other without conflict.

It should be noted that the terms "first," "second," and the like herein are merely used for distinguishing between different devices, modules, or units and not for limiting the order or interdependence of the functions performed by such devices, modules, or units.

It should be noted that references to "one", "a plurality" and "a plurality" in this disclosure are intended to be illustrative rather than limiting, and those skilled in the art will appreciate that "one or more" is intended to be construed as "one or more" unless the context clearly indicates otherwise.

The names of messages or information interacted between the devices of the present invention are for illustrative purposes only and are not intended to limit the scope of such messages or information.

The invention will be described in detail below with reference to the drawings in connection with embodiments.

As shown in FIG. 1, a flow chart of some embodiments of a trajectory prediction method according to a vehicle trajectory semantic analysis and deep belief network of the present invention is shown. The method for predicting the track according to the vehicle track semantic analysis and the deep belief network specifically comprises the following steps:

Step 101, acquiring a historical running track set of a target vehicle in a historical time period, wherein each historical running track in the historical running track set comprises at least one of the following: a target start point and a target end point.

In some embodiments, the execution subject of the trajectory prediction method is a background server, which may be, for example, a school bus management background server. On this basis, the execution subject may acquire a set of historical travel tracks of the target vehicle in a historical period from the vehicle track database. For example, after manually specifying the position information (longitude and latitude) of the target start point and the target end point, the execution subject may query the vehicle trajectory database for vehicle trajectory data including the target start point or the target end point in the history period and obtain a history travel trajectory set. The historical travel tracks in the historical travel track set may include a plurality of track points arranged in time, each track point represented by a triplet, each triplet including a longitude, latitude, and a timestamp. Wherein the target vehicle may be at least one school bus running on the same line. The historical time period may be a manually specified time period or a time period determined according to a certain condition.

102, Carrying out semantic analysis on each historical driving track in the historical driving track set to obtain a plurality of resident points corresponding to each historical driving track, wherein the plurality of resident points corresponding to each historical driving track in the historical driving track set respectively form a resident point set.

In some embodiments, the execution body may perform semantic analysis on each historical driving track. Specifically, semantic analysis can be performed by determining each historical driving track through the following steps, so that a plurality of residence points corresponding to each historical driving track are obtained:

First, for each historical driving track, determining whether a preset number of track points meeting a first preset condition exist in a track point sequence included in the historical driving track, wherein the first preset condition is that the preset number of track points are continuous track points, and the distance between any two adjacent track points in the preset number of track points is smaller than a preset distance threshold value.

Secondly, if the historical driving track exists, determining the resident points corresponding to the preset number of track points, and obtaining a plurality of resident points corresponding to each historical driving track. In the process, the resident points are identified based on the continuous track points meeting the first preset condition, so that more accurate resident point identification can be realized, and reasonable setting of the resident points is realized while the requirements of a school bus pick-up scene are met.

Specifically, a preset number of track points can be mapped into an electronic map to obtain a mapped electronic map, wherein the mapped electronic map comprises a plurality of track points corresponding to the preset number of track points one by one. And then, determining the minimum circumscribed circles corresponding to the preset number of mapping track points. Generally, at least three mapping track points in the preset number of mapping track points fall on the minimum circumscribing circle, and the rest mapping track points fall in the minimum circumscribing circle. It should be noted that, when two mapped track points falling on the minimum circumscribing circle have the same longitude or latitude, only two mapped track points may fall on the minimum circumscribing circle.

And then, determining a plurality of candidate residence positions which are positioned in the minimum circumscribed circle and meet preset residence conditions in the mapping electronic map. For example, the preset stay condition is that the candidate stay location is located at the arterial road. On the basis, for each candidate resident position in the plurality of candidate resident positions, determining the sum of the distances between the candidate resident position and the preset number of mapping track points, and obtaining the sum of the distances corresponding to each candidate resident position. And finally, selecting the smallest candidate residence position and the distance as residence points corresponding to the preset number of track points.

Step 103, determining average residence time length and residence frequency corresponding to residence points in the residence point set, sorting according to the average residence time length and residence frequency to obtain a residence point sequence, selecting a first number of residence points with front sorting from the residence point sequence, and determining the first number of residence points as candidate residence point sets.

In some embodiments, the executing entity may determine an average dwell time length and a dwell frequency corresponding to dwell points in the dwell point set. Specifically, a dwell point is selected from the dwell point set, and the following grouping operation is performed: determining whether at least one resident point with the distance smaller than the first distance exists in the resident point set, if so, determining the selected resident point and the at least one resident point as a resident point group, determining the ratio between the number of the resident points in the resident point group and the number of the historical driving tracks in the historical driving track set, determining the ratio as resident frequency, calculating the average value of resident time lengths corresponding to all the resident points in the resident point group, and obtaining average resident time length, wherein the average resident time length and the resident frequency corresponding to the selected resident point can be obtained. On this basis, the grouping operation described above may be continued for non-grouped dwells in the set of dwells, as needed, until all of the groups are grouped. In addition, if there is no at least one dwell point in the set of dwell points that is less than the first distance from the selected dwell point, the selected dwell points may be individually grouped.

On the basis, combining all resident points in the same resident point group in the grouped resident point set to obtain an updated resident point set. For example, one resident point is randomly reserved in the same resident point group, and the rest is deleted.

And sequencing all the resident points in the updated resident point set according to the corresponding average resident duration and resident frequency to obtain a resident point sequence. In practice, by grouping and merging the residence points in the residence point set, the residence points when different vehicles and different time points pass through similar positions (such as the same cell pick-up point) can be duplicated. The dwell time period corresponding to each dwell point can be determined by the time interval in which the plurality of continuous track points corresponding to the dwell point are located. The plurality of continuous track points corresponding to the residence point may be a preset number of track points, or may be more continuous track points than the preset number. The dwell time period corresponding to each dwell point may be a time difference between a first locus point and a last locus point in the plurality of continuous locus points. As an example, the average dwell time length and dwell frequency may be weighted and summed to obtain a dwell point score corresponding to each dwell point in the updated dwell point set. In practice, the higher the average dwell time, the higher the dwell frequency, the higher the dwell point score. On the basis, sorting all the resident points in the resident point set according to the order of the resident point scores from large to small, so as to obtain a resident point sequence. And then, selecting a first number of dwell points from the dwell point sequence, which are ranked first, namely selecting the first number of dwell points according to the order of the dwell point scores from large to small, and determining the first number of dwell points as a candidate dwell point set. Wherein the first number may be a pre-specified number, for example, may be 10.

Step 104, generating prompt information based on the candidate resident point set, and sending the prompt information to terminal devices respectively corresponding to the plurality of vehicle users, so that the plurality of vehicle users respectively select corresponding target resident points to obtain a plurality of target resident points and the number of vehicle users corresponding to each target resident point.

In some embodiments, as an example, the execution subject renders each resident point in the candidate resident point set in the electronic map, and combines the rendered electronic map and the preset prompt message text into the prompt message. For example, the hint text may be "please select your get-on location from a set of candidate stay points displayed in the electronic map". Further, the identification (for example, a mobile phone number or a communication account name) of the terminal device corresponding to each of the plurality of registered riding users may be acquired, and the prompt information may be transmitted to the terminal device corresponding to each of the plurality of riding users. On the basis, each riding user can select a corresponding target residence point through the corresponding terminal equipment, and therefore the terminal equipment can send the target residence point selected by each riding user to the execution main body. Therefore, the execution subject can count the number of passengers, the information of passengers and the like corresponding to each target residence point. Optionally, each riding user may further upload a face image as the reserved face image when selecting the target residence point.

And 105, planning paths according to the target starting point, the target ending point, the plurality of target residence points and the pre-configured road network data to obtain a plurality of planned paths.

In some embodiments, the executing entity may implement path planning by calling a path planning interface or adopting an open-source path planning algorithm, so as to obtain a plurality of planned paths, where each planned path includes a target starting point, a target ending point and a plurality of target residence points.

And 106, acquiring a real-time traffic flow thermodynamic diagram, and inputting the real-time traffic flow thermodynamic diagram into a deep belief network to obtain real-time road condition information.

In some embodiments, the executing entity may obtain the real-time traffic thermodynamic diagram through an interface provided by some electronic map applications, and input the real-time traffic thermodynamic diagram into a deep belief network to obtain real-time traffic information. The deep belief network can automatically learn high-order features in the image, and encode and decode through nodes in the hidden layer, so that more accurate real-time road condition information can be generated.

And 107, selecting a target planned path from the plurality of planned paths according to the real-time road condition information.

In some embodiments, the execution body may render the plurality of planned paths into a real-time traffic thermodynamic diagram, resulting in a rendered thermodynamic diagram. On the basis, intercepting an area with each planning path displayed in the rendering thermodynamic diagram to obtain an image block corresponding to each planning path; inputting the image blocks corresponding to each planning path into a convolutional neural network for feature extraction to obtain image block features; and after the characteristics of a plurality of image blocks corresponding to the plurality of planning paths are spliced with the real-time road condition information, inputting a path selection network to obtain path selection information, wherein the path selection information characterizes the number or the name of the selected path, and further, the selected path is determined to be a target planning path. Wherein the path selection network may be a pre-trained recurrent neural network. In practice, the training sample set can be utilized to train the cyclic neural network through machine learning algorithms such as back propagation, random gradient descent and the like, so that a path selection network is obtained. The training samples in the training sample set comprise a plurality of sample image block features, sample road condition information and manually marked target planning paths. The method comprises the steps of inputting the spliced sample image block characteristics and sample road condition information into a circulating neural network, taking a target planning path marked by people as expected output, calculating the difference between actual output and expected output by using a preset loss function, carrying out back propagation on the difference, and adjusting parameters of each layer of network, so that one iteration is completed. Similarly, multiple iterations may be performed until the target number of iterations is reached, thereby completing training and obtaining the path selection network.

And step 108, the target planning path is used as a predicted track to be sent to terminal equipment corresponding to a plurality of riding users respectively.

In some embodiments, semantic analysis is performed on the historical driving track set, so that a plurality of corresponding resident points are extracted, and then candidate resident points are determined, so that more accurate resident point identification and automatic extraction are realized, and the requirement of a school bus pick-up scene is met. In the process, the average residence time and residence frequency of the residence points are fully considered, so that reasonable planning of the residence points is realized, the residence points are not required to be set manually, and a foundation is laid for the application of automatic driving in a school bus pick-up scene. In addition, in the process of generating the predicted track, the depth belief network is used for extracting real-time traffic thermodynamic diagrams to obtain real-time road condition information, so that high-order features in the image can be extracted, and the path selection is assisted by the real-time road condition information, so that the predicted track is generated. In the process, the deep belief network can fully extract the high-order characteristics of the real-time traffic thermodynamic diagram, so that more accurate real-time road condition information is generated, and the generated prediction track is more accurate.

In some embodiments, in order to further solve the second technical problem described in the background section, that is, "in a school bus pick-up scenario, since a riding user is often an minor, how to guarantee the safety of the riding user by technical means is a problem to be solved, some embodiments of the present invention further include the following steps:

When the target vehicle is detected to reach each target resident point, acquiring face images of each actual riding user of the target resident point through an image acquisition device arranged at an upper door of the target vehicle, so as to obtain a plurality of actual face images;

And comparing the actual face images with the reserved face images of the riding users selecting the target residence to obtain a comparison result. Wherein, the comparison result represents whether the bus users which are not on the bus exist in a plurality of bus users which select the target residence point. If the comparison result represents that a plurality of bus users of the target residence point are selected to have bus users which are not on bus, determining the bus users which are not on bus as target bus users, judging whether the target riding user is an underage user according to the user information of the target riding user, and if the target riding user is the underage user, sending prompt information representing that the target riding user is not on the bus to a terminal corresponding to the associated user of the target riding user; the terminal equipment is also used for collecting face images of the riding users as reserved face images when the riding users select target residence points. Wherein the associated user of the target ride user may be an in-person user (e.g., guardian user) of the target ride user.

In the driving process of the target vehicle, acquiring images and voice data of a driver through an image acquisition device and a voice acquisition device which are arranged at a driving position of the target vehicle;

respectively extracting features of the driver image and the driver voice data to obtain image features and voice features;

And carrying out security assessment according to the image features and the voice features to obtain security assessment scores, and generating risk prompt information representing that the security risk exists if the security assessment scores are smaller than a preset score threshold value. The image features and the voice features can be input into a security evaluation network through security evaluation, and the security evaluation network is a logistic regression model. Optionally, after the risk prompt information is generated, braking treatment and emergency evacuation treatment can be performed on the target vehicle in time, wherein the emergency evacuation treatment comprises the steps of starting a vehicle door, starting a warning lamp, playing evacuation prompt voice and the like.

In addition, before the target planned path is sent to the terminal devices respectively corresponding to the plurality of riding users as the predicted track, the track prediction method further comprises the following steps:

And generating the prediction time for reaching each target residence point according to the real-time road condition information, the target planning path and the number of passengers corresponding to each target residence point. It was found in the study that the predicted time to reach each target dwell point has a correlation, i.e. the predicted time to reach a certain target dwell point is directly influenced by the predicted time of the target dwell point before that target dwell point, based on which a time-series prediction network (LSTM, long-short-term memory neural network) is used to generate the predicted time of each target dwell point. Specifically, real-time road condition information, a target planning path, the number of passengers corresponding to each target stay point and a stay point sequence are input into the time sequence prediction network, and the long-term memory neural network can fully utilize the previous information and apply the information to the current prediction, so that the generated prediction time reaching each target stay point is more accurate.

At this time, the target planned path is sent as a predicted trajectory to terminal devices respectively corresponding to a plurality of riding users, and the method includes:

And for each of the plurality of riding users, sending the predicted time and the target planning path corresponding to the target residence point selected by the riding user to the terminal equipment corresponding to the riding user.

In these embodiments, safety detection is performed during the passenger boarding phase, so that underage passengers who are not riding on time are discovered in time and their guardians are notified in time. In addition, in the running process of the vehicle, safety detection is carried out by collecting images of the driver and voice data of the driver, so that threat to safety of passengers caused by dangerous states of the driver is avoided.

In some embodiments, in order to further solve the third technical problem described in the background section, that is, "in the process of performing semantic analysis on a historical driving track set to obtain a resident point, there is a problem of redundancy of the resident point", in some embodiments of the present invention, performing semantic analysis on each historical driving track in the historical driving track set to obtain a plurality of resident points corresponding to each historical driving track, including the following substeps:

Determining whether a preset number of track points meeting a first preset condition exist in a track point sequence included in each historical driving track, wherein the first preset condition is that the preset number of track points are continuous track points, and the distance between any two adjacent track points in the preset number of track points is smaller than a preset distance threshold value; the specific value of the preset number may be set according to practical situations, for example, may be 5.

Step two, if the historical driving track exists, determining the resident points corresponding to the preset number of track points to obtain a plurality of resident points corresponding to each historical driving track;

determining a historical time interval corresponding to each resident point (namely a time interval in which a plurality of continuous track points corresponding to each resident point) for each resident point in a plurality of resident points corresponding to each historical driving track, and acquiring a road section average speed (namely an average speed of a plurality of vehicles driving on the road section) of a road section in which the resident point is located in the historical time interval;

Determining whether the average speed of the road section is lower than a preset speed and whether the average speed of the target vehicle in the historical time interval is matched with the average speed of the road section, if the average speed of the road section is lower than the preset speed and the average speed of the historical vehicle is matched with the average speed of the road section, deleting the resident point corresponding to the historical time interval from the plurality of resident points, and forming a residual resident point set by the residual at least one resident point;

Step five, for each resident point in the residual resident point set, acquiring historical driving state information of the target vehicle in a historical time interval corresponding to the resident point, and determining whether abnormal prompt information exists in the historical driving state information, wherein the abnormal prompt information represents that the target vehicle is in an abnormal operation state, the abnormal prompt information comprises the risk prompt information and fault prompt information, and the fault prompt information represents that the vehicle has a fault;

And step six, deleting the resident point from the rest resident point sets if the abnormal prompt information exists in the historical driving state information, and forming the resident point sets by at least one resident point.

It has been found that the problem of parking spot redundancy exists because factors due to traffic jams and vehicle anomalies are not considered in the identification process, so that slow vehicle speeds or vehicle parking caused by these factors are also identified as parking spots (i.e., pseudo-parking spots), thereby causing the parking spot redundancy. Based on this, in some embodiments, the problem of parking spot redundancy is solved by deleting the pseudo parking spots caused by these factors by considering the road section average vehicle speed and the historical driving state information.

The above description is only illustrative of the few preferred embodiments of the present invention and of the principles of the technology employed. It will be appreciated by persons skilled in the art that the scope of the invention referred to in the present invention is not limited to the specific combinations of the technical features described above, but also covers other technical features formed by any combination of the technical features described above or their equivalents without departing from the inventive concept described above. Such as the above-mentioned features and the technical features disclosed in the present invention (but not limited to) having similar functions are replaced with each other.

Claims (4)

1. The track prediction method based on the vehicle track semantic analysis and the deep belief network is characterized by comprising the following steps:

acquiring a historical driving track set of a target vehicle in a historical time period, wherein each historical driving track in the historical driving track set comprises at least one of the following: a target start point and a target end point;

Carrying out semantic analysis on each historical driving track in the historical driving track set to obtain a plurality of resident points corresponding to each historical driving track, wherein the resident points corresponding to each historical driving track in the historical driving track set form a resident point set;

Determining average residence time length and residence frequency corresponding to residence points in the residence point set, sorting according to the average residence time length and residence frequency to obtain a residence point sequence, selecting a first number of residence points with front sorting from the residence point sequence, and determining the first number of residence points as a candidate residence point set;

generating prompt information based on the candidate resident point set, and sending the prompt information to terminal equipment respectively corresponding to a plurality of vehicle users, so that the plurality of vehicle users respectively select corresponding target resident points to obtain a plurality of target resident points and the number of vehicle users corresponding to each target resident point;

Performing path planning according to the target starting point, the target ending point, the plurality of target residence points and the pre-configured road network data to obtain a plurality of planned paths;

acquiring a real-time traffic flow thermodynamic diagram and inputting the real-time traffic flow thermodynamic diagram into a deep belief network to obtain real-time road condition information;

Selecting a target planned path from the plurality of planned paths according to the real-time road condition information;

The target planning path is used as a predicted track to be sent to terminal equipment corresponding to the plurality of riding users respectively;

before the target planned path is sent to the terminal devices respectively corresponding to the plurality of riding users as the predicted track, the track prediction method further comprises the following steps:

Generating a prediction time for reaching each target residence point according to the real-time road condition information, the target planning path and the number of passengers corresponding to each target residence point;

And sending the target planned path as a predicted track to terminal devices respectively corresponding to the plurality of riding users, wherein the terminal devices comprise:

for each of the plurality of ride users, sending the predicted time and the target planned path corresponding to the target residence point selected by the ride user to a terminal device corresponding to the ride user;

The terminal equipment is also used for collecting a face image of the riding user as a reserved face image when the riding user selects a target residence point; the track prediction method further comprises the following steps:

When the target vehicle is detected to reach each target parking point, acquiring face images of each actual riding user of the target parking point through an image acquisition device arranged at an upper door of the target vehicle, so as to obtain a plurality of actual face images;

Comparing the actual face images with reserved face images of a plurality of riding users selecting the target residence to obtain a comparison result; if the comparison result representation selects a plurality of boarding users with the target residence, determining the boarding users with the target boarding users as the target boarding users, judging whether the target boarding users are underage users according to the user information of the target boarding users, and if the target boarding users are underage users, sending prompt information representing that the target boarding users are underage to the terminals corresponding to the associated users of the target boarding users.

2. The track prediction method according to claim 1, wherein the track prediction method further comprises:

in the driving process of the target vehicle, acquiring images of a driver and voice data of the driver through an image acquisition device and a voice acquisition device which are arranged at a driving position of the target vehicle;

Respectively carrying out feature extraction on the driver image and the driver voice data to obtain image features and voice features;

And carrying out security assessment according to the image features and the voice features to obtain security assessment scores, and generating risk prompt information representing that security risks exist if the security assessment scores are smaller than a preset score threshold value.

3. The track prediction method according to the semantic analysis and the depth belief network of the vehicle track according to claim 2, wherein each of the historical driving tracks includes a track point sequence, track points in the track point sequence are triplets, and the triplets include longitude, latitude and time stamps; and

Performing semantic analysis on each historical driving track in the historical driving track set to obtain a plurality of residence points corresponding to each historical driving track, wherein the semantic analysis comprises the following steps:

For each historical driving track, determining whether a preset number of track points meeting a first preset condition exist in a track point sequence included in the historical driving track, wherein the first preset condition is that the preset number of track points are continuous track points, and the distance between any two adjacent track points in the preset number of track points is smaller than a preset distance threshold;

If the historical driving track exists, determining the resident points corresponding to the preset number of track points, and obtaining a plurality of resident points corresponding to each historical driving track.

4. The track prediction method according to claim 3, wherein the determining the average residence time length and residence frequency corresponding to the residence points in the residence point set, and sorting according to the average residence time length and residence frequency, to obtain the residence point sequence, includes:

Selecting one resident point from the resident point set, determining whether at least one resident point with the distance smaller than a first distance exists in the resident point set, if so, determining the selected resident point and the at least one resident point as a resident point group, determining the ratio between the number of resident points in the resident point group and the number of historical driving tracks in the historical driving track set, determining the ratio as resident frequency, and calculating the average value of resident time lengths corresponding to all resident points in the resident point group to obtain average resident time length;

merging all resident points in the resident point group to obtain an updated resident point set;

And sequencing all the resident points in the updated resident point set according to the corresponding average resident duration and resident frequency to obtain a resident point sequence.