Detection and processing method for two-segment track cheating

Abstract

The invention discloses a detection and processing method for two-segment trajectory cheating, which identifies and repairs jumping, rollback and overlapping phenomena through a vector included angle cosine value and an ellipse method, thereby improving the integrity of trajectory data and ensuring the accurate calculation of mileage. In combination with trajectory clustering and a Markov chain model, the system can realize more accurate short-term and long-term trajectory prediction, and platform scheduling is optimized; by means of similarity analysis and unsupervised learning, the system can efficiently detect abnormal track behaviors, and automatic recognition can be achieved even under the condition that no annotated data exists; besides, through a transfer learning framework and in combination with track data of taxis and buses, the system can effectively detect illegal online taxi-hailing behaviors, and the problem of insufficient annotation data is solved; finally, the system accurately calculates the driving mileage of the vehicle through the repaired track, and the fairness and accuracy of cost calculation are ensured.

Description

Technical Field

[0001] This invention belongs to the field of ride-hailing trajectory fraud, specifically, it relates to a detection and processing method for two-segment trajectory fraud. Background Technology

[0002] With the rapid development of the ride-hailing industry, trajectory fraud has become a significant challenge for both platforms and users. Drivers falsify or tamper with driving trajectories, commonly employing methods such as trajectory jumps, rollbacks, and overlaps. These behaviors prevent platforms from accurately calculating actual mileage, affecting fair payments to users and increasing platform operating costs. Existing trajectory detection methods struggle to handle such complex fraudulent behaviors, especially when trajectories are discontinuous or abnormal. Current technologies cannot accurately repair and restore the true trajectory and have significant limitations in predicting future driving paths. Trajectory anomaly detection is inefficient due to the complexity of the data and the lack of labeled data, making it difficult to apply traditional supervised learning models. Furthermore, the detection of illegal ride-hailing vehicles relies on large amounts of labeled data, but the limited availability of such data makes traditional supervised learning methods ineffective in identifying such behaviors, leading to platform oversight failures and an increase in potential illegal activities. Finally, existing billing systems often fail to accurately calculate the vehicle's actual mileage when faced with trajectory jumps, overlaps, and other anomalies, resulting in inaccurate fare calculations that negatively impact user experience and the platform's pricing fairness.

[0003] In view of this, the present invention is proposed. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide a detection and processing method for cheating on bi-segmented trajectories, thereby solving the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the basic concept of the technical solution adopted by the present invention is as follows: A method for detecting and handling cheating using bi-segmented trajectories includes the following steps: S1: Receive and process the trajectory data uploaded by the vehicle and divide the trajectory data into a series of discrete coordinate points. Analyze the integrity of the trajectory, detect the cosine value of the vector angle between adjacent trajectory points, and determine whether there is a jump phenomenon. S2: After detecting the jump phenomenon in S1, the ellipse method is used to calculate the overlapping part between the two trajectories. By identifying the overlapping points, the jump and overlapping parts are removed, and the remaining trajectory is repaired. S3: After the S2 repair is completed, the trajectory data is clustered and combined with the Markov chain model to predict the future driving path, so as to achieve short-term and long-term trajectory prediction. S4: After S3 is completed, perform similarity analysis on the repaired trajectory, detect abnormal trajectory behavior, and combine unsupervised learning to identify abnormal trajectories; S5: After identifying abnormal trajectories in S4, the system uses a transfer learning framework, combined with trajectory data from taxis and buses, to detect whether there are any illegal ride-hailing activities. S6: Based on the trajectory data repaired by S5, recalculate the vehicle's mileage, output detection results and prediction information, and perform corresponding cost calculations.

[0006] Optionally, when determining whether a bisegmented trajectory exists, the process first involves receiving trajectory data uploaded from the vehicle. This trajectory data consists of continuous coordinate points and includes both temporal and spatial information. Next, the acquired trajectory data is segmented into discrete coordinate points, forming a sequence of trajectory points. Each trajectory point contains the vehicle's position coordinates at a specific moment. Then, using spatial and temporal vector information, the angle between adjacent points is calculated to determine the continuity of the trajectory. When abnormal changes occur in the trajectory, a formula is used to... Calculate the cosine of the angle between the vectors of the trajectory points to determine whether the directions of motion of the trajectory points are consistent. A value less than 0 indicates a significant jump or regression in the trajectory. When the calculated cosine value is less than 0, it confirms an abnormal regression or jump in the trajectory, suggesting that the anomaly is caused by external interference or cheating. and It is the vector formed by adjacent trajectory points.

[0007] Optionally, after detecting a jump phenomenon, the formula is used... Calculate the overlap between the two trajectories, where, and These are the major and minor axes of the ellipse, respectively. and The coordinates of the trajectory points are used. Then, calculations are performed on each trajectory point. If a point in the second trajectory segment falls within the ellipse, it is identified as an overlapping point. Subsequently, after identifying overlapping points, trajectory points in the jump and overlapping parts are removed. After removing the jump and overlapping parts, the remaining trajectory points are considered valid trajectory data. Then, after removing invalid points, the system reconnects the remaining trajectory points. Finally, the repaired trajectory generates a real trajectory without false points by calculating the distance and direction between adjacent points.

[0008] Optionally, after trajectory repair is completed in S3, the trajectory data will be clustered and automatically classified. Similar vehicle trajectories will be grouped together. After trajectory clustering, each trajectory point is considered a state in a Markov chain, representing the vehicle's position at a specific time point. Next, the position points in the trajectory data are mapped to a discrete set of states. Subsequently, based on historical trajectory data, the Markov chain analyzes the frequency of vehicle transitions between states to construct a transition probability matrix. Each item in this matrix represents the probability of a vehicle transitioning from one state to another. Finally, the Markov chain model predicts the probability of a vehicle moving from one location to another using multiple time steps.

[0009] Optionally, when predicting future travel paths, the Markov chain model can determine whether to make short-term or long-term predictions by defining the number of time steps: For short-term trajectory prediction, the Markov chain uses the current state and the state transition matrix PPP to predict the vehicle's trajectory over a short period of time. By applying the transition matrix multiple times, the system predicts the vehicle's possible positions in the next few time steps. in, Let PPP represent the state vector at the current moment, and let PPP be the transition matrix. For long-term predictions, the system uses multi-step transitions of a Markov chain to predict the vehicle's trajectory over a longer period. By recursively applying the state transition matrix, it predicts even further into the future. By using a multi-step transition matrix, the system can predict driving paths over a longer period of time.

[0010] Optionally, after trajectory repair is completed, the system performs similarity analysis on the trajectory data, using Fourier transform and wavelet transform to convert the trajectory data from the time domain to the frequency domain. Then, it uses formulas... The system calculates a similarity metric between trajectory points to determine if any trajectories significantly deviate from the normal path. Then, by minimizing the distance from each trajectory point to the cluster center of the clustered trajectory data, normal trajectories are identified. Trajectory points deviating from the cluster center are marked as abnormal trajectories. The system identifies potentially abnormal trajectories by comparing the distances of each trajectory point to the cluster center. After completing similarity analysis and unsupervised learning, the detected abnormal trajectory behaviors are labeled and classified, and the detection results are output. This is the distance function between two trajectory points.

[0011] Optionally, taxi and bus trajectory data are used. A preliminary classifier is then trained using this data. Based on features such as trajectory time intervals, stops, and acceleration, a model capable of recognizing trajectory patterns is built. After training the taxi and bus models, transfer learning is used to apply these learned trajectory features to the trajectory data of ordinary vehicles. A convolutional neural network is then used to further extract deeper features from the trajectory data. By extracting high-order features from vehicle trajectory points and path changes, the system identifies whether ordinary vehicles conform to public transportation trajectory patterns. Finally, by analyzing the trajectories of ordinary vehicles, vehicles that significantly deviate from the trajectory patterns of legal taxis and buses are detected. If a vehicle's trajectory features differ significantly from legal trajectories, the system marks the vehicle as a potential illegal ride-hailing vehicle. After detecting potential illegal ride-hailing vehicles, the detection results are output, and these vehicles are marked for further analysis and law enforcement tracking.

[0012] Optionally, in step S6, when calculating the cost, firstly, using the formula... The system calculates the distance between adjacent coordinate points in the trajectory and accumulates the distances of all segments to obtain the total mileage. Next, it removes abnormal jumps or overlapping trajectory points during the calculation. Then, based on the prediction results of a Markov chain model, it helps the platform or passengers estimate future trips. Finally, based on the corrected mileage, the system recalculates the vehicle's trip cost. Finally, the system outputs the final cost result to both the passenger and the platform. and These are the coordinates of the trajectory points.

[0013] By adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art. Of course, any product implementing the present invention does not necessarily need to achieve all of the following advantages at the same time: This invention identifies and corrects jumps, backtracking, and overlaps by using the cosine of the vector angle and the elliptic method, thereby improving the integrity of trajectory data and ensuring accurate calculation of mileage. Combining trajectory clustering and Markov chain models, the system can achieve more accurate short-term and long-term trajectory prediction, optimizing platform scheduling. Utilizing similarity analysis and unsupervised learning, the system can efficiently detect abnormal trajectory behavior, automatically identifying it even without labeled data. Furthermore, through a transfer learning framework combined with trajectory data from taxis and buses, the system can effectively detect illegal ride-hailing behavior, solving the problem of insufficient labeled data. Finally, the system accurately calculates vehicle mileage using the corrected trajectory, ensuring the fairness and accuracy of fare calculation.

[0014] The specific embodiments of the present invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0015] The accompanying drawings described below are merely some embodiments. Those skilled in the art can obtain other drawings based on these drawings without any creative effort. In the drawings: Figure 1 A flowchart illustrating the detection and handling methods for cheating on bi-segmented trajectories; Figure 2 This is one of the schematic diagrams of the experimental process in Example 1; Figure 3 This is the second schematic diagram of the experimental process in Example 1; Figure 4 This is the third schematic diagram of the experimental process in Example 1; Figure 5 This is the fourth schematic diagram of the experimental process in Example 1; Figure 6 This is the fifth schematic diagram of the experimental process in Example 1; Figure 7 This is the sixth schematic diagram of the experimental process in Example 1; Figure 8 This is one of the schematic diagrams of the experimental process in Example 2; Figure 9 This is the second schematic diagram of the experimental process in Example 2; Figure 10 This is the third schematic diagram of the experimental process in Example 2; Figure 11 This is the fourth schematic diagram of the experimental process in Example 2.

[0016] It should be noted that these accompanying drawings and textual descriptions are not intended to limit the scope of the invention in any way, but rather to illustrate the concept of the invention to those skilled in the art by referring to specific embodiments. Detailed Implementation

[0017] The invention will now be described in further detail with reference to the accompanying drawings.

[0018] Please see Figure 1-11 As shown, this embodiment provides a method for detecting and processing cheating on bi-segmented trajectories, including the following steps: S1: Receive and process the trajectory data uploaded by the vehicle and divide the trajectory data into a series of discrete coordinate points. Analyze the integrity of the trajectory, detect the cosine value of the vector angle between adjacent trajectory points, and determine whether there is a jump phenomenon. S2: After detecting the jump phenomenon in S1, the ellipse method is used to calculate the overlapping part between the two trajectories. By identifying the overlapping points, the jump and overlapping parts are removed, and the remaining trajectory is repaired. S3: After the S2 repair is completed, the trajectory data is clustered and combined with the Markov chain model to predict the future driving path, so as to achieve short-term and long-term trajectory prediction. S4: After S3 is completed, perform similarity analysis on the repaired trajectory, detect abnormal trajectory behavior, and combine unsupervised learning to identify abnormal trajectories; S5: After identifying abnormal trajectories in S4, the system uses a transfer learning framework, combined with trajectory data from taxis and buses, to detect whether there are any illegal ride-hailing activities. S6: Based on the trajectory data repaired by S5, recalculate the vehicle's mileage, output detection results and prediction information, and perform corresponding cost calculations.

[0019] This invention identifies and corrects jumps, backtracking, and overlaps by using the cosine of the vector angle and the elliptic method, thereby improving the integrity of trajectory data and ensuring accurate calculation of mileage. Combining trajectory clustering and Markov chain models, the system can achieve more accurate short-term and long-term trajectory prediction, optimizing platform scheduling. Utilizing similarity analysis and unsupervised learning, the system can efficiently detect abnormal trajectory behavior, automatically identifying it even without labeled data. Furthermore, through a transfer learning framework combined with trajectory data from taxis and buses, the system can effectively detect illegal ride-hailing activities, solving the problem of insufficient labeled data. Finally, the system accurately calculates vehicle mileage using the corrected trajectory, ensuring the fairness and accuracy of fare calculation. In this embodiment, when determining whether a bi-segmented trajectory exists, firstly, trajectory data uploaded by the vehicle is received. This trajectory data consists of continuous coordinate points and includes both temporal and spatial information. Next, the acquired trajectory data is segmented into discrete coordinate points, forming a sequence of trajectory points. Each trajectory point contains the vehicle's position coordinates at a specific moment. Then, using spatial and temporal vector information, the angle between adjacent points is calculated to determine the continuity of the trajectory. When abnormal changes occur in the trajectory, a formula is used to... Calculate the cosine of the angle between the vectors of the trajectory points to determine whether the directions of motion of the trajectory points are consistent. A value less than 0 indicates a significant jump or regression in the trajectory. When the calculated cosine value is less than 0, it confirms an abnormal regression or jump in the trajectory, suggesting that the anomaly is caused by external interference or cheating. and It is a vector formed by adjacent trajectory points. By calculating the cosine of the angle between the vectors of adjacent trajectory points, this embodiment can accurately identify jumps or backtracking phenomena in the trajectory, alerting potential cheating behaviors, thereby ensuring the authenticity of the trajectory and providing a basis for subsequent trajectory repair, avoiding the problem of false mileage calculation caused by abnormal trajectories.

[0020] In this embodiment, after detecting the jump phenomenon, the formula is used... Calculate the overlap between the two trajectories, where, and These are the major and minor axes of the ellipse, respectively. and The coordinates of the trajectory points are used, and then calculations are performed on each trajectory point. If a point in the second trajectory segment falls within the ellipse, it is identified as an overlapping point. Subsequently, after identifying overlapping points, trajectory points in the jump and overlapping parts are removed. After removing jump and overlapping parts, the remaining trajectory points are considered valid trajectory data. Then, after removing invalid points, the system reconnects the remaining trajectory points. Finally, the repaired trajectory generates a true trajectory without false points by calculating the distance and direction between adjacent points. This embodiment uses the ellipse method to calculate the overlapping parts in the trajectory, identifies and removes invalid jump and overlapping points, and reconnects the remaining trajectory points to ensure the integrity and continuity of the trajectory, laying the foundation for accurately calculating the vehicle's mileage.

[0021] In this embodiment, after trajectory repair is completed in S3, the trajectory data is clustered and automatically classified. Similar vehicle trajectories are grouped together. After trajectory clustering, each trajectory point is considered a state in a Markov chain, representing the vehicle's position at a specific time point. Next, the position points in the trajectory data are mapped to a discrete set of states. Subsequently, based on historical trajectory data, the Markov chain analyzes the frequency of vehicle transitions between states to construct a transition probability matrix. Each item in this matrix represents the probability of a vehicle transitioning from one state to another. Finally, the Markov chain model predicts the probability of a vehicle moving from one location to another using multiple time steps. By clustering the trajectory data, this embodiment effectively groups similar trajectories and analyzes state transition frequencies using a Markov chain model, enabling short-term and long-term path prediction for vehicles and optimizing the platform's vehicle scheduling decision-making capabilities.

[0022] In this embodiment, the Markov chain model predicts future travel paths by defining the number of time steps, which determines whether to make short-term or long-term predictions. For short-term trajectory prediction, the Markov chain uses the current state and the state transition matrix PPP to predict the vehicle's trajectory over a short period of time. By applying the transition matrix multiple times, the system predicts the vehicle's possible positions in the next few time steps. in, Let PPP represent the state vector at the current moment, and let PPP be the transition matrix. For long-term predictions, the system uses multi-step transitions of a Markov chain to predict the vehicle's trajectory over a longer period. By recursively applying the state transition matrix, it predicts even further into the future. By using a multi-step transition matrix, the system can predict driving paths over a longer period. This embodiment employs Fourier transform and wavelet transform to convert trajectory data into the frequency domain, and uses unsupervised learning to identify abnormal trajectories that deviate from the normal path, greatly improving the accuracy and efficiency of abnormal trajectory detection. It can be effectively applied even without labeled data.

[0023] In this embodiment, after trajectory repair is completed, the system performs similarity analysis on the trajectory data, using Fourier transform and wavelet transform to convert the trajectory data from the time domain to the frequency domain. Then, it uses the formula... The system calculates a similarity metric between trajectory points to determine if any trajectories significantly deviate from the normal path. Then, by minimizing the distance from each trajectory point to the cluster center of the clustered trajectory data, normal trajectories are identified. Trajectory points deviating from the cluster center are marked as abnormal trajectories. The system identifies potentially abnormal trajectories by comparing the distances of each trajectory point to the cluster center. After completing similarity analysis and unsupervised learning, the detected abnormal trajectory behaviors are labeled and classified, and the detection results are output. Distance function between two trajectory points In this embodiment, taxi and bus trajectory data are used to train a preliminary classifier. Then, based on features such as trajectory time intervals, stop points, and acceleration, a model capable of recognizing trajectory patterns is built. After training the taxi and bus models, transfer learning is used to apply these learned trajectory features to the trajectory data of ordinary vehicles. A convolutional neural network is used to further extract deep features from the trajectory data. The system extracts high-order features such as vehicle trajectory points and path changes to identify whether ordinary vehicles conform to public transportation trajectory patterns. Finally, by analyzing the trajectories of ordinary vehicles, vehicles that significantly deviate from the trajectory patterns of legal taxis and buses are detected. If a vehicle's trajectory features differ significantly from legal trajectories, the system marks the vehicle as a possible illegal ride-hailing vehicle. After detecting possible illegal ride-hailing vehicles, the detection results are output, and these vehicles are marked for further analysis and law enforcement tracking. This embodiment utilizes taxi and bus trajectory data for model training through transfer learning and applies convolutional neural networks to extract deep features, effectively identifying illegal ride-hailing behavior that deviates from legal public transportation trajectory patterns, improving the reliability and accuracy of detection.

[0024] In this embodiment, when calculating the cost in step S6, firstly, the formula is used... The system calculates the distance between adjacent coordinate points in the trajectory and accumulates the distances of all segments to obtain the total mileage. Next, it removes abnormal jumps or overlapping trajectory points during the calculation. Then, based on the prediction results of a Markov chain model, it helps the platform or passengers estimate future trips. Finally, based on the corrected mileage, the system recalculates the vehicle's trip cost. Finally, the system outputs the final cost result to both the passenger and the platform. and These are the coordinates of the trajectory points. In this embodiment, after removing jump and overlapping trajectory points, the actual mileage of the vehicle is recalculated, and the cost is calculated by combining the prediction results of the Markov chain model, ensuring the transparency and accuracy of the cost calculation and providing fair billing results for passengers and the platform.

[0025] Example 1: Extensive data analysis of this type of cheating reveals the following patterns. Halfway through trajectory 1, the driver's uploaded location suddenly returned to a position previously visited on trajectory 1, and then caught up to the driver's current location. See... Figure 2-3 The leftmost line represents trajectory 1, the middle line represents trajectory 2, and the rightmost line represents the final merged state of the two trajectories, with the green part on the rightmost line representing the overlapping portion.

[0026] For the first case in the analysis, the overlapping portion of the bisegmental trajectory can be calculated. Removing the overlapping portion yields a complete trajectory. (The following is a more detailed explanation.) Figure 4 It can be seen that the effect after the repair is significant. First, we obtain the estimated mileage by taking the driver's origin and destination and selecting the same route. Then, we compare the mileage after the driver cheated with the mileage after the repair using this method. Calculate the overlapping portion of the bisegmental trajectory A trajectory is formed by connecting coordinate points. The key to finding overlapping parts is locating the starting and ending points of these overlapping points. This patent uses the ellipse method, which involves constructing ellipses for every pair of coordinate points on one trajectory. Then, points on the other trajectory are used to determine which points fall within the range of a certain ellipse. If a point falls within a certain ellipse, it is considered an overlapping point. This process is used to find all overlapping points on the other trajectory. Figure 5 As shown. Figure 6 As shown, by substituting the coordinates of the points into the ellipse formula, if a point on another trajectory falls within the ellipse trajectory, then that point is considered a coincident point. .

[0027] like Figure 7-8 As shown, Track 1 first uploads location points according to the real track. When it is halfway through, Track 1 starts to simulate a track according to the cheating software. Then Track 2 reverts to the location point where the simulated track started and then uploads the normal track points.

[0028] The leftmost line shows the travel path of trajectory 1, the middle line shows the travel path of trajectory 2, and the rightmost line shows the trajectory after the two trajectories have been processed and merged by the program. It can be seen that the two trajectories have a backtracking phenomenon and some overlap.

[0029] Example 2: like Figure 7-8 As shown, Track 1 first uploads location points according to the real track. When it is halfway through, Track 1 starts to simulate a track according to the cheating software. Then Track 2 reverts to the location point where the simulated track started and then uploads the normal track points.

[0030] The leftmost line shows the travel path of trajectory 1, the middle line shows the travel path of trajectory 2, and the rightmost line shows the trajectory after the two trajectories have been processed and merged by the program. It can be seen that the two trajectories have a backtracking phenomenon and some overlap.

[0031] like Figure 9 As shown, the existence of a backtracking phenomenon can be determined by calculating the cosine of the vector formed by the two points at the end of trajectory 1 and the vector formed by the end of trajectory 1 and the starting point of trajectory 2. Then, it calculates whether there is a backtracking phenomenon between the two segmented trajectories. This involves calculating the local vector of the endpoint of trajectory 1, and then calculating the vectors of the endpoint of trajectory 1 and the starting point of trajectory 2. The cosine value of these two vectors is then checked to see if it is less than 0. If the cosine value is less than 0, the cheating scenario is considered to exist. Next, it calculates whether there is an overlap between the two trajectories. If so, the cheating scenario is considered an overlapping backtracking scenario. The overlap can be determined by the cosine value of the vectors between trajectory points, as detailed below. Figure 10 As shown, The cosine value ranges from -1 to 1. When vectors A and B are in the same direction, the cosine value is 1. If the cosine value of a vector is greater than a certain threshold (which can be a value close to 1), we consider that there are overlapping trajectory points, and thus find all overlapping points.

[0032] To address cheating in overlapping rollback scenarios, the system considers Track 1 to contain a portion of the real trajectory and a portion of the simulated trajectory. The simulated portion, which is the trajectory formed from the start point of the overlapping part of the two trajectories to the end point of Track 1, is discarded. This allows the system to reconstruct the entire real trajectory, as shown below. Figure 11 As shown.

[0033] This invention is not limited to the embodiments described above. Anyone should understand that structural changes made under the guidance of this invention, and any technical solutions that are the same as or similar to this invention, fall within the protection scope of this invention. Technical aspects, shapes, and structures not described in detail in this invention are all publicly known technologies.

Claims

1. A method for detecting and processing cheating using bi-segmented trajectories, characterized in that, Includes the following steps: S1: Receive and process the trajectory data uploaded by the vehicle and divide the trajectory data into a series of discrete coordinate points. Analyze the integrity of the trajectory, detect the cosine value of the vector angle between adjacent trajectory points, and determine whether there is a jump phenomenon. S2: After detecting the jump phenomenon in S1, the ellipse method is used to calculate the overlapping part between the two trajectories. By identifying the overlapping points, the jump and overlapping parts are removed, and the remaining trajectory is repaired. S3: After the S2 repair is completed, the trajectory data is clustered and combined with the Markov chain model to predict the future driving path, so as to achieve short-term and long-term trajectory prediction. S4: After S3 is completed, perform similarity analysis on the repaired trajectory, detect abnormal trajectory behavior, and combine unsupervised learning to identify abnormal trajectories; S5: After identifying abnormal trajectories in S4, the system uses a transfer learning framework, combined with trajectory data from taxis and buses, to detect whether there are any illegal ride-hailing activities. S6: Based on the trajectory data repaired by S5, recalculate the vehicle's mileage, output detection results and prediction information, and perform corresponding cost calculations.

2. The method for detecting and processing cheating on bi-segmented trajectories according to claim 1, characterized in that, To determine the presence of a bisegmented trajectory, the system first receives trajectory data uploaded from the vehicle. This trajectory data consists of continuous coordinate points and contains both temporal and spatial information. Next, the acquired trajectory data is segmented into discrete coordinate points, forming a sequence of trajectory points. Each trajectory point contains the vehicle's position coordinates at a specific moment. Then, using spatial and temporal vector information, the angle between the vectors of adjacent points is calculated to determine the continuity of the trajectory. When abnormal changes occur in the trajectory, a formula is used to detect them. Calculate the cosine of the angle between the vectors between trajectory points to determine whether the movement directions of the trajectory points are consistent. If cos(θ) is less than 0, it indicates that there is an obvious jump or back phenomenon in the trajectory. When the calculated cosine value is less than 0, it is determined that there is an abnormal back or jump in the trajectory, indicating that the abnormality of this segment of the trajectory is caused by external interference or cheating. Here, A and B are the vectors formed by adjacent trajectory points.

3. The method for detecting and processing cheating on bi-segmented trajectories according to claim 1, characterized in that, After detecting the jump phenomenon, the formula is used. The overlapping portion between two trajectory segments is calculated, where a and b are the major and minor axes of the ellipse, respectively, and x and y are the coordinates of the trajectory points. Then, for each trajectory point, calculations are performed. If a point in the second trajectory segment falls within the ellipse's range, that point is identified as an overlapping point. Subsequently, after identifying overlapping points, trajectory points in jumps and overlapping portions are removed. After removing jumps and overlapping portions, the remaining trajectory points are considered valid trajectory data. Then, after removing invalid points, the system reconnects the remaining trajectory points. Finally, the repaired trajectory generates a true trajectory without false points by calculating the distance and direction between adjacent points.

4. The method for detecting and processing cheating on bi-segmented trajectories according to claim 1, characterized in that, After trajectory repair is completed in S3, the trajectory data will be clustered and automatically classified. Similar vehicle trajectories will be grouped together. After trajectory clustering, each trajectory point is regarded as a state in a Markov chain, and the state represents the vehicle's position at a specific time point. Then, the position points in the trajectory data are mapped to a discrete set of states. Subsequently, based on historical trajectory data, the Markov chain analyzes the frequency of vehicle transitions between states to construct a transition probability matrix. Each item in this matrix represents the probability of the vehicle transitioning from one state to another. Finally, the Markov chain model predicts the probability of the vehicle moving from one position to another using multiple time steps.

5. The method for detecting and processing cheating on bi-segmented trajectories according to claim 1, characterized in that, When Markov chain models predict future travel paths, they can determine whether to make short-term or long-term predictions by defining the number of time steps. For short-term trajectory prediction, the Markov chain uses the current state and the state transition matrix PPP to predict the vehicle's trajectory over a short period of time. By applying the transition matrix multiple times, the system predicts the vehicle's position at several time steps in the future. t+1 =P·S t Among them, S t Let PPP represent the state vector at the current moment, and let PPP be the transition matrix. For long-term predictions, the system uses multi-step transitions of a Markov chain to predict the vehicle's trajectory in future time intervals. By recursively applying the state transition matrix, it predicts even further into the future. t+n =P n ·S t By using a multi-step transition matrix, the system can predict driving paths over a longer period of time.

6. The method for detecting and processing cheating on bi-segmented trajectories according to claim 1, characterized in that, After trajectory repair is completed, the system performs similarity analysis on the trajectory data, using Fourier transform and wavelet transform to convert the trajectory data from the time domain to the frequency domain. Then, it uses formulas... The system calculates the similarity metric between trajectory points to determine if any trajectory significantly deviates from the normal path. Then, by minimizing the distance from the trajectory point to the cluster center of the clustered trajectory data, normal trajectories are identified. Trajectory points that deviate from the cluster center are marked as abnormal trajectories. The system identifies abnormal trajectories by comparing the distance of each trajectory point to the cluster center. After completing the similarity analysis and unsupervised learning, the detected abnormal trajectory behaviors are labeled and classified, and the detection results are output. Here, d(T1, T2) is the distance function between two trajectory points.

7. The method for detecting and processing cheating on bi-segmented trajectories according to claim 1, characterized in that, The system first uses taxi and bus trajectory data. Then, it trains a preliminary classifier using this data. Next, based on trajectory time intervals, stop points, and acceleration features, it builds a model capable of recognizing trajectory patterns. After training the taxi and bus models, transfer learning is used to apply these learned trajectory features to the trajectory data of ordinary vehicles. A convolutional neural network is then used to further extract deeper features from the trajectory data. By extracting high-order features from vehicle trajectory points and path changes, the system identifies whether ordinary vehicles conform to public transportation trajectory patterns. Finally, by analyzing the trajectories of ordinary vehicles, it detects vehicles that significantly deviate from the legal taxi and bus trajectory patterns. If a vehicle's trajectory features differ significantly from legal trajectories, the system marks the vehicle as an illegal ride-hailing vehicle. After detecting illegal ride-hailing vehicles, the detection results are output, and these vehicles are marked for further analysis and law enforcement tracking.

8. The method for detecting and processing cheating on bi-segmented trajectories according to claim 1, characterized in that, In step S6, when calculating the cost, firstly, using the formula... The system calculates the distance between adjacent coordinates in the trajectory and accumulates the distances of all segments to obtain the total mileage. Next, it removes abnormal jumps or overlapping trajectory points during the calculation. Then, based on the prediction results of a Markov chain model, it helps the platform or passengers estimate future trips. Finally, based on the corrected mileage, the system recalculates the vehicle's trip cost. Finally, the system outputs the final cost result to the passenger and the platform, where x... i and y i These are the coordinates of the trajectory points.

Images