A morphological perception-based riding road curve fitting method

Abstract

The application discloses a riding road curve fitting method based on shape perception, first extracts intersection trajectories and road section trajectories according to historical riding trajectory data and intersection positions and their influence ranges, and obtains turning trajectory clusters in intersections and road section trajectory clusters between intersections based on a trajectory clustering algorithm, then models shape features of each trajectory segment in the trajectory cluster based on directions and positions of trajectory points to obtain a shape feature vector of the trajectory segment, then uses an XGBoost classification model to predict a shape category of each trajectory in the trajectory cluster, and counts the number of trajectories belonging to each shape in the trajectory cluster to obtain a shape of the trajectory cluster, and finally selects a corresponding optimal center line fitting method for each type of trajectory cluster to perform fitting to obtain a road curve.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of trajectory mining, and particularly relates to a riding road curve fitting method based on shape perception. BACKGROUND

[0002] With the large-scale increase in the number of non-motor vehicles, non-motor vehicle-related services have developed rapidly and entered a period of rapid growth. In order to reach the destination conveniently and quickly, people often use non-motor vehicles as a means of transportation. Due to the lack of professional and accurate non-motor vehicle riding navigation map, it often enters the non-motor vehicle restricted area when riding. In addition, it often happens that the destination is reached by relying on the route provided by the two-wheeled vehicle navigation software which is not updated in time, and the events such as entering deep mountains and forest areas occur. The construction and timely update of the non-motor vehicle riding map can ensure efficient riding route planning and improve the riding experience of non-motor vehicles. In addition, the rapid growth of non-motor vehicle-related services has accumulated a large amount of non-motor vehicle riding trajectory data, providing a data basis for the inference of the riding map. However, due to the diversity of the actual road shape and the randomness of the riding behavior, the riding trajectories on roads of different shapes are obviously different, making the fitting of the riding road curve face serious challenges. Therefore, it is urgent to design a riding road curve fitting method based on shape perception.

[0003] At present, the method of road curve fitting in the map inference algorithm based on cross-linking can be divided into two basic types. One method is based on trajectory segment, which uses the influence range of the intersection or "binds" the trajectory with the intersection in the intersection recognition process, so as to screen out the trajectory cluster between adjacent intersections, and then use a single center line fitting method to fit the road curve. However, this method is limited by the shape of the trajectory, and currently there is no center line fitting method suitable for fitting road curves of all shapes. The other is a method based on trajectory points, which searches along the exit or entry direction of the intersection, and uses the first intersection found as the adjacent intersection. In the search process, a sliding window model is used to select the most suitable GPS position from all trajectory points as the shape point of the road curve. However, due to the skew distribution of the trajectory and the randomness of the riding behavior, the detection effect of the road curve in the sparse area of the trajectory is poor. SUMMARY

[0004] The technical problem to be solved by the present application is to overcome the shortcomings of the prior art, and to provide a riding road curve fitting method based on shape perception, which can effectively and accurately fit the riding road curve using non-motor vehicle trajectory data, and provide effective decision support for navigation services and path planning.

[0005] To achieve the above technical purpose, the technical scheme adopted by the present application is as follows:

[0006] A riding road curve fitting method based on shape perception, comprising the following steps:

[0007] Step S1: Extract intersection trajectories and road segment trajectories according to historical riding trajectory data and intersection positions and their influence ranges, and obtain turning trajectory clusters in intersections and road segment trajectory clusters between intersections based on a trajectory clustering algorithm.

[0008] Step S2: Model the shape features of each trajectory segment in the trajectory cluster based on the direction and position of the trajectory point.

[0009] Step S3: Use an XGBoost classification model to predict the shape category of each trajectory in the trajectory cluster, and count the number of trajectories belonging to each shape in the trajectory cluster to obtain the shape of the trajectory cluster.

[0010] Step S4: Based on the shape of the trajectory cluster obtained in step S3, select the corresponding optimal centerline fitting method for each shape trajectory cluster to obtain the road curve.

[0011] To optimize the above technical solution, the specific measures taken also include:

[0012] The above step S1 specifically includes:

[0013] Step 1.1) Segment each historical riding trajectory according to the intersection position and its influence range to obtain intersection trajectory segments located in the intersection and road segment trajectory segments located between intersections.

[0014] Step 1.2) For intersection trajectory segments, use Fréchet distance to measure the similarity between trajectories, and consider that trajectories with adjacent road segments but inconsistent headings cannot be distinguished by Fréchet distance. Therefore, the direction weight is included in the similarity measurement, so the similarity measurement formula between trajectories is as follows:

[0015] S(tr a ,tr b )=d Fréchet (tr a ,tr b )+w(tr a ,tr b )*ζ

[0016]

[0017] Where d Fréchet (tr a ,tr b ) represents the Fréchet distance between trajectories tr a and tr b , w(tr a ,tr b ) represents the direction weight between trajectories tra and tr b between the direction of tr a and tr b , the weight is set to 1 when the angle difference between the start (or end) point of tr

[0018]

[0019] where β is determined by specific experiments (set to 7 in the present application), and n is the number of trajectories in the corresponding intersection. Finally, based on the results of the similarity measure, the DBSCAN clustering algorithm is used to cluster the trajectories to obtain the turning trajectory cluster, and the important parameter eps is determined by ζ, and minSample is set to 3 in the present application.

[0020] Step 1.3) For road segment trajectory segments, clustering is performed according to the direction consistency of the trajectories, i.e. trajectory segments with the same starting intersection and arrival intersection are clustered into the same road segment trajectory cluster.

[0021] The above step S2 specifically includes:

[0022] Considering that the trajectories passing through different forms of roads have obvious differences in form, the form of the trajectory can be divided into 6 categories, which are: straight trajectory segment, gentle turning trajectory segment, right-angle turning trajectory segment, sharp turning trajectory segment, U-shaped turning trajectory segment and self-intersecting trajectory segment. In order to distinguish the 6 forms of trajectories, the following features are extracted as the form features of the trajectories: 1) direction difference (f a ), i.e. the direction difference between the first point and the last point of the trajectory segment; 2) length (f l ), i.e. the length of the trajectory; 3) displacement length ratio (f dl ), i.e. the ratio of the distance between the two end points of the trajectory segment to the length of the trajectory.

[0023] The above step S3 specifically includes:

[0024] Based on the form features obtained in step S2, a 3-dimensional feature vector tr.feature = [f a , f l , f dl ] is constructed for each trajectory segment, and then an XGBoost classification model is used to construct a multi-classifier to predict the form category of each trajectory. Finally, for each trajectory segment cluster, the number of trajectory segments belonging to each form category is calculated, and the form with the most number of trajectory segments is regarded as the predicted form of the cluster.

[0025] The above step S4 specifically includes:

[0026] Considering that a single center line fitting method is not suitable for fitting all morphological trajectory clusters, the application proposes to select the corresponding optimal center line fitting method for fitting different morphological trajectory clusters. For trajectory clusters in a straight line form, the Sweeping Line algorithm is selected to fit them. This method scans along a fixed direction and is difficult to adapt to trajectory clusters with sudden changes in direction, so it is mainly suitable for straight-line motion. For trajectory clusters in the forms of gentle turning, right-angle turning and sharp turning, the k-section main curve algorithm is selected to fit them. This method aims to generate a smooth curve that passes through the trajectory point cluster from the middle. It performs well in fitting trajectory clusters in the forms of gentle turning, right-angle turning and sharp turning, but performs poorly in the case of trajectory clusters with unclear forms, such as self-intersecting trajectory clusters. For trajectory clusters in the form of U-shaped turning, the B-spline curve algorithm is selected to fit them. This method models the road curve as a B-spline curve, which is suitable for trajectories with continuous and long turning forms. For trajectory clusters in the form of self-intersection, the trajectory sampling algorithm is selected to fit them. This method selects a representative trajectory segment from the trajectory cluster as the road curve, which is more suitable for some complex trajectories, i.e., self-intersecting trajectories. However, due to its strong dependence on the quality of the sampled trajectory segment, the fitting result may be too short or deviate from the center.

[0027] The application has the following beneficial effects:

[0028] 1. In view of the diversification of trajectory forms, the application first proposes a center line fitting strategy based on form perception. The form of a trajectory cluster is perceived according to the form characteristics of the trajectories in the trajectory cluster, and the optimal center line fitting method is selected for trajectory clusters in different forms.

[0029] 2. The application uses real trajectory data to experimentally evaluate the road curve fitting method, and the experimental results verify the effectiveness of the method. BRIEF DESCRIPTION OF DRAWINGS

[0030] Figure 1 The figure is a framework diagram of the application, a riding road curve fitting method based on form perception.

[0031] Figure 2 The figure is a road network effect diagram generated by the Huang method in the embodiment of the application.

[0032] Figure 3 The figure is a road network effect diagram generated by the method of the application in the embodiment of the application. DETAILED DESCRIPTION

[0033] The application will be further described in detail in combination with the following specific embodiments and drawings. The process, conditions, experimental methods, etc. for implementing the application are the general knowledge and common sense in the art, and the application does not have special restrictions.

[0034] The application discloses a riding road curve fitting method based on shape perception, as shown in the accompanying drawings. Figure 1 As shown in the accompanying drawings, first, intersection trajectories and road section trajectories are extracted according to historical riding trajectory data and intersection positions and their influence ranges, and turning trajectory clusters in intersections and road section trajectory clusters between intersections are obtained based on a trajectory clustering algorithm, then shape features of each trajectory section in the trajectory cluster are modeled based on the direction and position of the trajectory point to obtain a shape feature vector of the trajectory section, then an XGBoost classification model is used to predict the shape category of each trajectory in the trajectory cluster, and the number of trajectories belonging to each shape in the trajectory cluster is counted to obtain the shape of the trajectory cluster, finally, an optimal center line fitting method corresponding to each type of trajectory cluster is selected for fitting to obtain a road curve.

[0035] As shown in the accompanying drawings, Figure 1 The application discloses a riding road curve fitting method based on shape perception, which comprises the following steps:

[0036] Step S1: according to historical riding trajectory data and intersection positions and their influence ranges, intersection trajectories and road section trajectories are extracted, and turning trajectory clusters in intersections and road section trajectory clusters between intersections are obtained based on a trajectory clustering algorithm.

[0037] In the embodiment, step S1 specifically comprises:

[0038] Step 1.1) each historical riding trajectory is segmented according to intersection positions and their influence ranges to obtain intersection trajectory sections located in intersections and road section trajectory sections located between intersections.

[0039] Step 1.2) for the intersection trajectory sections, the similarity between trajectories is measured by using Fréchet distance, and in view of the problem that trajectories with adjacent road sections but inconsistent directions cannot be distinguished by Fréchet distance, a direction weight is introduced into the similarity measurement, so that the similarity measurement formula between trajectories is as follows:

[0040] S(tr a ,tr b )=d Fréchet (tr a ,tr b )+w(tr a ,tr b )*ζ

[0041]

[0042] Wherein, d Fréchet (tr a ,tr b ) represents the Fréchet distance between trajectories tr a and tr b , w(tr a ,trb ) represents the direction weight between trajectories tr a and tr b , which is set to 1 when the angle difference between the start (or end) points of tr a and tr b is greater than 15°, otherwise 0. ζ is an adaptive similarity threshold, which is calculated as follows:

[0043]

[0044] where β is determined by specific experiments (set to 7 in the present application), and n is the number of trajectories in the corresponding intersection. Finally, based on the results of the similarity measure, the DBSCAN clustering algorithm is used to cluster the trajectories to obtain the turning trajectory cluster, with the important parameter eps determined by ζ, and minSample set to 3 in the present application.

[0045] Step 1.3) For road segment trajectory segments, clustering is performed based on the direction consistency of the trajectories, i.e. trajectory segments with the same starting intersection and arrival intersection are clustered into the same road segment trajectory cluster.

[0046] Step S2: Model the shape features of each trajectory segment in the trajectory cluster based on the direction and position of the trajectory points.

[0047] In the embodiment, step S2 specifically includes:

[0048] Considering that the shapes of trajectories passing through different types of roads are significantly different, the shape of a trajectory can be divided into 6 categories, namely: straight trajectory segment, gentle turning trajectory segment, right-angle turning trajectory segment, sharp turning trajectory segment, U-shaped turning trajectory segment, and self-intersecting trajectory segment. To distinguish the 6 shapes of the trajectory, the following features are extracted as the shape features of the trajectory: 1) direction difference (f a ), i.e. the direction difference between the first point and the last point of the trajectory segment; 2) length (f l ), i.e. the length of the trajectory; 3) displacement length ratio (f dl ), i.e. the ratio of the distance between the two end points of the trajectory segment to the length of the trajectory.

[0049] Step S3: Use an XGBoost classification model to predict the shape category of each trajectory in the trajectory cluster, and count the number of trajectories belonging to each shape in the trajectory cluster to obtain the shape of the trajectory cluster.

[0050] In the embodiment, step S3 specifically includes:

[0051] Based on the shape features obtained in step S2, a 3-dimensional feature vector tr.feature = [f a , f l , f dl] and then a multi-classifier is built using the XGBoost classification model to predict the morphological category of each trajectory. Finally, for each trajectory segment cluster, the number of trajectory segments belonging to each morphological category is calculated, and the morphological category with the most trajectory segments is considered as the predicted morphological category of the cluster.

[0052] Step S4: Based on the morphologies of the trajectory clusters obtained in step S3, select the corresponding optimal center line fitting method for each morphological trajectory cluster to perform fitting to obtain the road curve.

[0053] In the embodiment, step S4 specifically comprises:

[0054] Considering that a single center line fitting method is not suitable for fitting all morphological trajectory clusters, the present application proposes to select the corresponding optimal center line fitting method for fitting different morphological trajectory clusters. For a straight morphological trajectory cluster, the Sweeping Line algorithm is selected to fit it, which scans along a fixed direction and is difficult to adapt to trajectory clusters with sudden direction changes, and therefore is mainly suitable for straight motion. For a gentle turning, right angle turning and sharp turning morphological cluster, the k-section principal curve algorithm is selected to fit it, which aims to generate a smooth curve passing through the trajectory point cluster from the middle, and performs well in fitting gentle turning, right angle turning and sharp turning morphologies, but performs poorly in the case of ambiguous trajectory cluster morphology, such as self-intersecting trajectory clusters. For a U-turn morphological cluster, the B-spline curve algorithm is selected to fit it, which models the road curve as a B-spline curve, which is suitable for trajectories with continuous and long turning morphologies. For a self-intersecting morphological cluster, the trajectory sampling algorithm is selected to fit it, which selects a representative trajectory segment from the trajectory cluster as the road curve, which is more suitable for some complex morphological trajectories, i.e. self-intersecting trajectories, but the fitting result may be too short or deviate from the center due to its strong dependence on the quality of the sampled trajectory segment.

[0055] To verify the effectiveness of the present application, real non-motor vehicle trajectory data and corresponding intersection information of the riding road curve fitting are selected, and based on the public riding road network, the existing map inference method is compared and evaluated. The selected comparison method is Huang, which is a cross-link based map inference method, and the part of the road curve fitting in this method is used as a comparison, as shown in Figure 2 Fig. 3 shows the road network effect diagram generated by the huang method near the intersection, Figure 3The road network effect picture generated in the same area based on the method is shown, which can accurately identify the turning path in the two-way road and the intersection, and can provide more effective help for navigation broadcast and path planning. In quantitative evaluation, the inferred map is matched with the real map within a certain distance threshold d by using the graph sampling evaluation method, the evaluation indexes are quantitatively evaluated by using Precision (precision), Recall (recall) and F1-score, and the comparison evaluation results are shown in Table 1. The experimental results show that the riding map obtained by using the method for fitting the riding road curve is obviously better than the effect obtained by using the prior art method for inference.

[0056] Table 1 Comparison table of effects of different methods

[0057]

[0058] In summary, the present application comprehensively uses DBSCAN clustering technology, sweeping line fitting technology, B-spline fitting technology, K-section main curve technology, trajectory sampling technology and other technologies, forms a riding road curve fitting method based on shape perception, can accurately fit the road curve through trajectory data mining, and thus provides effective help for navigation broadcast and path planning.

[0059] The above is only the preferred embodiment of the present application, and the protection scope of the present application is not limited to the above-mentioned embodiments. Any technical solution falling within the idea of the present application shall fall within the protection scope of the present application. It should be noted that, for ordinary skilled persons in the art, some improvements without departing from the principle of the present application shall be regarded as the protection scope of the present application.

Claims

1. A morphological perception based curve fitting method for cycling paths, characterized in that, The method comprises the following steps: Step S1: extracting intersection trajectory and road segment trajectory according to historical riding trajectory data and intersection position and its influence range, and obtaining intersection turning trajectory cluster and road segment trajectory cluster between intersections based on trajectory clustering algorithm; The step S1 specifically comprises: Step 1.1) segmenting each historical riding trajectory according to intersection position and its influence range to obtain intersection trajectory segment located in intersection and road segment trajectory segment located between intersections; Step 1.2) for intersection trajectory segment, using Fréchet distance to measure the similarity between trajectories, and incorporating direction weight into similarity measurement, the similarity measurement formula between trajectories is as follows: S(tr a ,tr b ) = d Fréchet (tr a ,tr b ) + w(tr a ,tr b )*ζ where d Fréchet (tr a ,tr b ) represents the Fréchet distance between trajectories tr a and tr b , w(tr a ,tr b ) represents the direction weight between trajectories tr a and tr b , which is set to 1 when the angle difference of the start or end point of tr a and tr b is greater than 15°, otherwise 0; ζ is the adaptive similarity threshold, which is calculated as follows: Wherein, β is determined by specific experiment parameter tuning, and n is the number of trajectories in the corresponding intersection; Based on the results of similarity measurement, DBSCAN clustering algorithm is used to cluster trajectories to obtain turning trajectory cluster, and the parameter eps is determined by ζ, and minSample is set to 3; Step 1.3) for road segment trajectory segment, clustering trajectory segments with the same starting intersection and arrival intersection to the same road segment trajectory cluster according to the consistency of trajectory direction; Step S2: modeling the shape feature of each trajectory segment in the trajectory cluster based on the direction and position of trajectory points; Step S3: using XGBoost classification model to predict the shape category of each trajectory in the trajectory cluster, and counting the number of trajectories belonging to each shape in the trajectory cluster to obtain the shape of the trajectory cluster; Step S4: based on the shape of the trajectory cluster obtained in step S3, selecting the corresponding optimal center line fitting method for each shape trajectory cluster to fit to obtain the road curve; The step S4 specifically comprises: For trajectory clusters of different shapes, the corresponding optimal center line fitting method is selected for fitting; for straight shape trajectory cluster, sweeping line algorithm is selected for fitting; for gentle turning, right angle turning and sharp turning shape cluster, k segment main curve algorithm is selected for fitting; for U-shaped turning shape cluster, B-spline curve algorithm is selected for fitting; for self-intersection shape cluster, trajectory sampling algorithm is selected for fitting.

2. The morphological perception based riding road curve fitting method according to claim 1, wherein, The step S2 specifically comprises: Trajectories passing through different shape roads are divided into: straight trajectory segment, gentle turning trajectory segment, right angle turning trajectory segment, sharp turning trajectory segment, U-shaped turning trajectory segment and self-intersection trajectory segment; in order to distinguish the shape of trajectory, the following features are extracted as the shape features of trajectory: 1) Direction difference f a i.e. the direction difference of the first and last points of the trajectory segment; 2) length f l i.e. the length of the trajectory; 3) Displacement length ratio f dl i.e. the ratio of the distance between the two end points of a trajectory segment to the length of the trajectory.

3. The morphological perception based riding road curve fitting method according to claim 1, wherein, The step S3 specifically comprises: Based on the morphological features obtained in step S2, a 3-dimensional feature vector tr.feature = [f a ,f l ,f dll ] is constructed for each trajectory segment, and then an XGBoost classification model is used to construct a multi-classifier to predict the morphological category of each trajectory; finally, for each trajectory segment cluster, the number of trajectory segments belonging to each morphological category is calculated, and the morphological category with the most trajectory segments is regarded as the predicted morphological category of the cluster.

Images