Shared bicycle dangerous riding behavior detection method based on image recognition

Abstract

The invention discloses a shared bicycle dangerous riding behavior detection method based on image recognition, and the method comprises the steps: obtaining riding track data with a timestamp, matching the riding track data with a high-precision map, carrying out the preprocessing and feature extraction of a matched image, marking track feature parameters, making a training set and a verification set according to a proportion, and constructing and initializing a detection neural network. Training by using the training set, verifying and evaluating by using the verification set, and finally acquiring to-be-identified trajectory data, matching into an image, preprocessing and inputting into a neural network to obtain an identification result. The method comprises a track data acquisition unit, a map matching unit and the like, a timestamp is accurate to a second level, a high-precision map can present various geographic information, a dangerous riding track covers retrograde driving, road crossing and illegal left turning, the ratio of a training set to a test set is 7: 3, and network evaluation comprises two evaluation indexes of accuracy and recall rate. The dangerous riding behavior of the shared bicycle can be effectively detected, and riding safety and traffic order are guaranteed.

Description

Technical Field

[0001] The present invention belongs to the field of smart transportation and artificial intelligence technology, and specifically relates to a method for detecting dangerous riding behaviors of shared bicycles based on image recognition. Background Art

[0002] The traditional way of cycling safety supervision mainly relies on manual inspections and on-site law enforcement. This method has problems such as low efficiency, limited coverage, and high cost, and it is difficult to effectively curb the occurrence of dangerous riding behaviors of shared bicycles. The shared bicycle dangerous riding behavior detection method based on smart transportation and artificial intelligence technology can obtain the shared bicycle riding trajectory data, match the motion trajectory with the map base map, determine whether the rider has dangerous riding behavior, and output the detection results. This method has the advantages of high efficiency, accuracy, and wide coverage. It can significantly improve the efficiency and effectiveness of cycling safety supervision and provide strong support for urban traffic management. Therefore, the development of an efficient and intelligent shared bicycle dangerous riding behavior detection method is of great significance to improving cycling safety and maintaining traffic order. Summary of the invention

[0003] To solve the above problems, the present invention discloses a method for detecting dangerous riding behaviors of shared bicycles based on image recognition, which aims to accurately detect dangerous behaviors in shared bicycle riding and improve riding safety and management efficiency.

[0004] To achieve the above object, the technical solution of the present invention is as follows: A method for detecting dangerous riding behaviors of shared bicycles based on image recognition comprises the following steps: S1. Obtain the shared bicycle riding trajectory data with timestamp; S2. Match the trajectory data to the high-precision map; S3. Preprocess the matched trajectory riding image and extract trajectory feature parameters; S4. Marking the extracted trajectory feature parameters as normal riding and dangerous riding trajectories; S5. Create training sets and validation sets according to the feature parameters in proportion; S6. Construct a shared bicycle dangerous riding behavior detection neural network and initialize its network parameters to obtain a primary shared bicycle dangerous riding behavior detection neural network; S7. Train the primary shared bicycle dangerous riding behavior detection neural network through the training set; S8. Put the validation set into the trained shared bicycle dangerous riding behavior detection neural network for verification and evaluate the network; S9. Collect the trajectory data of the shared bicycle to be identified and match it with the high-precision map to form an image to be identified, preprocess the image to be identified, and input the preprocessed image to be identified into the completed shared bicycle dangerous riding behavior detection neural network to obtain the shared bicycle dangerous riding behavior recognition result.

[0005] Preferably, in S1, the shared bicycle riding trajectory data with timestamps has timestamps accurate to the second level, and each timestamp is t1, t2, t3...tn with the same interval time, which can accurately reflect the time sequence of riding.

[0006] Preferably, in S2, Qgis software is used to match the trajectory data to a high-precision map using a matching algorithm.

[0007] Preferably, in S3, the preprocessing of the matched trajectory riding image refers to normalizing the image size to ensure the consistency of the image size during subsequent analysis and processing, and the characteristic parameter is the trajectory riding direction.

[0008] Preferably, in S4, the dangerous riding trajectory includes riding in the wrong direction, making an illegal left turn, and crossing the road.

[0009] Preferably, in S5, the ratio of the training set to the test set is 7:3.

[0010] Preferably, in S6, the network parameter initialization operation uses a random initialization method to initialize the weight parameters and bias parameters in the neural network.

[0011] Preferably, in S7, the specific training process includes the following steps: S21: Normalize all data in the training set; S22: The processed data is fed into the neural network and processed by the convolutional layer, pooling layer and fully connected layer in sequence; S23: Calculate the loss function of the trained neural network, and adjust the neural network based on the principle of minimizing the loss function.

[0012] Preferably, in S8, the evaluation index for evaluating the network includes accuracy and recall, and the formula for accuracy is: , the recall formula is: .

[0013] The beneficial effects of the present invention are: (1) The present invention achieves the effect of accurately detecting dangerous behaviors in shared bicycle riding and improving riding safety and management efficiency by adopting a series of measures, including obtaining shared bicycle riding trajectory data with timestamps accurate to the second level and with the same interval time, matching it to a high-precision map, and then preprocessing the matched trajectory riding images, extracting trajectory feature parameters, marking normal and dangerous riding trajectories, making training sets and verification sets, and constructing and training neural networks.

[0014] (2) We also normalized the image size to ensure image size consistency, clarified dangerous riding trajectories including riding in the wrong direction, illegal left turn, and crossing the road, set the ratio of training set to test set to 7:3, used random initialization method to initialize the neural network parameters, and used a detailed training process and accuracy and recall rate as evaluation indicators to further optimize the detection method and ensure the accuracy and stability of the detection.

[0015] (3) Compared with the prior art, the present invention integrates a variety of technical means to form a complete shared bicycle dangerous riding behavior detection system based on image recognition, which improves the accuracy and comprehensiveness of the detection of shared bicycle dangerous riding behavior, can more efficiently manage shared bicycle riding, and greatly improve riding safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic diagram of the process of the present invention.

[0017] Figure 2 It is a structural block diagram of the present invention.

[0018] Figure 3 This is a schematic diagram of the dangerous riding trajectory described in the present invention.

[0019] Figure 4 This is a schematic diagram of a normal riding trajectory according to the present invention. DETAILED DESCRIPTION

[0020] The present invention will be further explained below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention.

[0021] The core idea of ​​the present invention is to analyze the abnormal riding state characteristics of shared bicycle users when three dangerous riding behaviors such as riding against the flow, crossing the road and turning left illegally occur, study the occurrence law of dangerous riding states, match images based on shared bicycle trajectory points and high-precision maps, mark normal and dangerous riding behaviors, input the training set into the model for training, test the trained model with the test set, extract on-the-way dangerous riding behavior events in different traffic scenarios, identify the evolution law of riding risk state characteristics, and study the architecture of dangerous riding behavior recognition system in multiple traffic scenarios.

[0022] like Figure 1 As shown, the method for detecting dangerous riding behaviors of shared bicycles based on image recognition according to the present invention comprises the following steps: S1. With the help of the shared bicycle's own positioning system or other relevant data collection devices, the riding trajectory data with timestamps accurate to the second level is obtained. The timestamp sequence is t1; t2; t3...tn. These timestamps have the same interval and can accurately reflect the order and duration of riding, building an accurate time information foundation for subsequent analysis.

[0023] S2. Select a high-precision map that can present detailed geographic information including roads, intersections, landmarks, etc., and use a matching algorithm to accurately match each coordinate point and its corresponding timestamp in the acquired shared bicycle riding trajectory data to the corresponding position on the high-precision map, to ensure the accurate landing point of the trajectory data on the map, and to enhance the correlation and accuracy between the trajectory and the map.

[0024] S3. Perform pre-processing operations such as size normalization on the matched trajectory riding images to ensure that the image size is uniform during subsequent processing. On this basis, extract key trajectory feature parameters such as riding direction, which will serve as an important basis for judging whether the riding behavior is dangerous.

[0025] S4. Classify and label the extracted trajectory feature parameters into normal and dangerous riding trajectories. The feature parameters of normal riding are consistent with the conventional riding mode, while the feature parameters of dangerous riding trajectories include sudden changes in riding direction and trajectory direction that does not comply with traffic regulations, including three dangerous riding behaviors: riding against the flow, illegal left turn, and crossing the road. This labeling provides a classification basis for subsequent data set construction and neural network training.

[0026] S5. Divide the feature parameters into a training set and a validation set in a ratio of 7:3. The training set is used to train the neural network to grasp the feature rules, and the validation set is used to evaluate the performance of the trained network.

[0027] S6. Create a neural network specifically for detecting dangerous riding behaviors of shared bicycles, and initialize its network parameters. Use a random initialization method to initialize the weight parameters, bias parameters, etc. in the neural network to form a primary shared bicycle dangerous riding behavior detection neural network with a basic structure and initial parameter settings.

[0028] S7. Training the primary shared bicycle dangerous riding behavior detection neural network through the training set: specifically including the following steps: S71: Normalize all data in the training set: Normalize the data in the training set to unify the feature parameters in different value ranges into a standard range to prevent the training effect of the neural network from being affected by the large difference in the data value range.

[0029] S72: The processed data is put into the neural network and processed in turn through the convolution layer, pooling layer and fully connected layer: the normalized data is input into the neural network, and the data is processed in turn through the convolution layer to extract local features, the pooling layer to reduce the data dimension and retain the key features, and the fully connected layer to comprehensively process the features extracted by the previous layers.

[0030] S73: Calculate the loss function for the trained neural network and adjust the neural network based on the principle of minimizing the loss function: Calculate the loss function based on the output of the neural network and the actual marking result (normal or dangerous riding trajectory). Based on the principle of minimizing the loss function, use back propagation algorithm and other methods to adjust the convolution kernel weights of the neural network and the connection weight parameters of the fully connected layer, so as to continuously optimize the neural network during the training process.

[0031] S8. Input the validation set into the trained shared bicycle dangerous riding behavior detection neural network for verification. The network predicts and outputs based on the validation set data, and then evaluates the network through evaluation indicators such as accuracy and recall. The accuracy calculation formula is: (TP + TN) / (TP + TN + FP + FN) (TP is a true positive example, TN is a true negative example, FP is a false positive example, and FN is a false negative example); the recall calculation formula is: TP / (TP + FN). Further optimize the network performance based on the evaluation results to ensure that the network can accurately detect dangerous riding behaviors.

[0032] S9. Collect the trajectory data of the shared bicycle to be identified, match it with the high-precision map to generate the image to be identified, pre-process the image to be identified, and then input the pre-processed image into the trained and verified shared bicycle dangerous riding behavior detection neural network. The network analyzes and judges based on the input image features, and outputs the identification result of whether there is dangerous riding behavior in the image to be identified, thereby realizing the identification of dangerous riding behavior of shared bicycles.

[0033] Figure 2 This is a structural block diagram of a method for detecting dangerous riding behaviors of shared bicycles based on image recognition. This figure shows the architecture of a method for detecting abnormal riding behaviors of shared bicycles, which consists of multiple units. The following is a detailed explanation of each unit: Track data acquisition unit: corresponds to step S1. Through the shared bicycle's own positioning system or other data acquisition devices, the riding track data with time stamps t1, t2, t3...tn accurate to the second level are obtained. These time stamps are at the same interval and can accurately reflect the riding order and duration.

[0034] Map matching unit: corresponds to step S2. A high-precision map is selected, and the coordinate points and their timestamps in the acquired riding trajectory data are accurately matched with the corresponding positions on the map using a matching algorithm.

[0035] Image preprocessing and feature extraction unit: corresponds to step S3. Perform preprocessing operations such as size normalization on the matched trajectory riding image to ensure uniform image size. Extract key trajectory feature parameters such as riding direction as an important basis for judging whether riding behavior is dangerous.

[0036] Feature labeling unit: corresponds to step S4. The extracted trajectory feature parameters are classified and labeled into normal and dangerous riding trajectories. Normal riding features conform to the conventional mode, and dangerous riding features include three behaviors: driving against the flow, turning left illegally, and crossing the road, which provide a classification basis for subsequent data set production and neural network training.

[0037] Dataset preparation unit: corresponds to step S5. The feature parameters are divided into a training set and a validation set in a ratio of 7:3.

[0038] Neural network construction and initialization unit: corresponds to step S6. Create a neural network specifically for detecting dangerous riding behaviors of shared bicycles, and use random initialization and other methods to initialize the weights and bias parameters to obtain a primary detection neural network.

[0039] Model training unit: corresponds to step S7. It includes three sub-steps: S71: Normalize the data in the training set to unify the feature parameters with different value ranges into a standard range to prevent the large difference in data value ranges from affecting the training effect.

[0040] S72: The normalized data is input into the neural network, and the convolution layer is used to extract local features, the pooling layer is used to reduce the data dimension and retain the key features, and the fully connected layer is used to comprehensively process the features extracted by the previous layers.

[0041] S73: Calculate the loss function based on the output results of the neural network and the actual marking results. According to the principle of minimizing the loss function, use the back propagation algorithm to adjust the neural network parameters to continuously optimize the network.

[0042] Model verification and evaluation unit: corresponds to step S8. The verification set is input into the trained neural network for verification. The network predicts the output based on the verification set data. The network is then evaluated using two evaluation indicators: accuracy ((TP + TN) / (TP + TN + FP + FN)) and recall (TP / (TP + FN)). The network performance is optimized based on the evaluation results.

[0043] Behavior recognition unit: corresponds to step S9. Collect the track data of the shared bicycle to be identified, match it with the high-precision map to form the image to be identified, pre-process it and input it into the trained and verified neural network. The network analyzes and judges based on the image features, and outputs the identification result of whether there is dangerous riding behavior, so as to realize the identification of dangerous riding behavior of shared bicycles.

[0044] Figure 3 , Figure 4 A schematic diagram of dangerous and normal riding trajectories of a shared bicycle dangerous riding behavior detection method based on image recognition involved in the present invention is shown.

[0045] The figure clearly shows the normal and abnormal riding trajectories of shared bicycles in road environments. The purpose is to accurately distinguish and mark these two different types of riding trajectories so that they can be subsequently input into the convolutional neural network model for in-depth analysis and processing. For these input image data, the convolutional neural network can efficiently extract key "feature" information from the image with its own convolution and pooling layer structure characteristics, and then further process the "compressed image information" through the subsequent fully connected layer, and finally output the corresponding results.

[0046] The base map of the map should be a high-precision satellite map, which shows in detail the urban environment elements such as roads, green belts and buildings. At the same time, the map has obvious road signs and various transportation facilities. This setting can very intuitively show the riding trajectory of shared bicycles and provide a clear visual reference for accurately judging riding behavior. It should be noted that Figure 3 , Figure 4 It is only a schematic diagram of dangerous and normal riding trajectories. Its main purpose is to show the difference between dangerous riding trajectories and normal riding trajectories of shared bicycles.

[0047] Below Figure 3 , Figure 4 Expand detailed description: The two positions in the figure are marked as the starting and ending points of the shared bicycle focus key trajectory, and the direction of the bicycle icon is the riding direction of the shared bicycle. Figure 3 It includes three dangerous riding trajectories: riding against the flow, crossing the road, and turning left illegally. Figure 4 For comparison with the corresponding normal trajectory.

[0048] From these two diagrams, we can see that the criteria for judging whether a shared bicycle is in a dangerous riding trajectory are: Going against the flow: The displayed track is in the form of a straight line, and the riding direction is the direction indicated by the bicycle icon. The cyclist maintains a relatively stable driving direction on the road, but does not ride in the direction of the road sign marked on the map, which indicates that the cyclist is going against the flow on this section of the road.

[0049] The corresponding normal riding trajectory characteristics are that the rider can always maintain a stable driving direction on the corresponding road and ride strictly in accordance with the direction of the road signs marked on the map.

[0050] Crossing the road: The continuous points of the trajectory obviously do not cross the sidewalk area marked on the map, which indicates that the cyclist did not cross the road from the sidewalk according to normal traffic rules, but chose to cross the road.

[0051] The corresponding normal riding trajectory feature is that the continuous points of the shared bicycle's trajectory clearly pass through the sidewalk area marked on the map, indicating that the rider completes the action of crossing the road from the sidewalk.

[0052] Illegal left turn: The trajectory shows that when the cyclist made a left turn at the intersection, although the angle changed, the change was relatively gradual. In addition, the cyclist did not drive in a straight line for a while at the intersection to wait for the opportunity to turn left as required by normal traffic rules, but turned left directly at the intersection.

[0053] The corresponding normal riding trajectory feature is that when the cyclist arrives at the intersection and makes a left turn, the trajectory shows an obvious angle change, and before making the left turn, the cyclist first drives a straight distance on the road section corresponding to the traffic light, and then completes the left turn in accordance with traffic regulations and intersection traffic conditions.

[0054] In the technical solution of the present invention, a shared bicycle dangerous riding behavior detection model built using convolutional neural network technology is used as an example to explain. The detection model is mainly composed of three core parts, namely the convolution layer, the pooling layer and the fully connected layer, each of which plays a different and critical role in the entire detection process.

[0055] The main function of the convolution layer is to accurately and efficiently extract the local and global features of the image from the input image data. By performing sliding convolution operations on the image through the convolution kernel, the feature information of different regions can be captured, laying a solid foundation for subsequent analysis and processing. The main purpose of the pooling layer is to effectively process the data processed by the convolution layer and greatly reduce the parameter magnitude of the data. By adopting common pooling methods such as maximum pooling or average pooling, the dimension of the data can be reduced and the features can be further purified while retaining the key feature information, thereby improving the computational efficiency and generalization ability of the entire model. The fully connected layer is mainly responsible for in-depth processing of the "compressed image information" obtained after processing by the pooling layer, and finally outputs the corresponding detection results. It establishes a fully connected relationship between each neuron, conducts a comprehensive analysis and judgment of the feature information, and completes the accurate judgment of the shared bicycle riding behavior corresponding to the input image.

[0056] In the actual detection process, the shared bicycle track image, which has been accurately matched with the high-precision map, is accurately input into the shared bicycle dangerous riding behavior detection model built above. Based on its internal complex and precise calculation mechanism, the model conducts comprehensive and in-depth analysis and detection of the shared bicycle riding track corresponding to the input image data.

[0057] After precise analysis and detection by the model, three different types of dangerous riding behaviors were successfully identified from this shared bicycle riding trajectory, namely, riding in the wrong direction, crossing the road, and illegal left turns.

[0058] The method for detecting dangerous riding behaviors of shared bicycles based on image recognition described in the present invention can achieve the following detection indicators:

[0059] The accuracy and recall data presented above clearly reflect the accuracy of the detection model in identifying and detecting different dangerous riding behaviors, and provide an important basis for evaluating the effectiveness and reliability of the shared bicycle dangerous riding behavior detection method involved in the present invention.

[0060] It should be noted that the above content only illustrates the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. For ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications all fall within the protection scope of the claims of the present invention.

Claims

1. A method for detecting dangerous riding behaviors of shared bicycles based on image recognition, characterized in that: The following steps are involved: S1. Obtain the shared bicycle riding trajectory data with timestamp; S2. Match the trajectory data to the high-precision map; S3. Preprocess the matched trajectory riding image and extract trajectory feature parameters; S4. Marking the extracted trajectory feature parameters as normal riding and dangerous riding trajectories; S5. Create training sets and validation sets according to the feature parameters in proportion; S6. Construct a shared bicycle dangerous riding behavior detection neural network and initialize its network parameters to obtain a primary shared bicycle dangerous riding behavior detection neural network; S7. Training the primary shared bicycle dangerous riding behavior detection neural network through the training set; S8. Put the validation set into the trained shared bicycle dangerous riding behavior detection neural network for verification, and evaluate the network; S9. Collect the trajectory data of the shared bicycle to be identified and match it with the high-precision map to form an image to be identified, preprocess the image to be identified, and input the preprocessed image to be identified into the completed shared bicycle dangerous riding behavior detection neural network to obtain the shared bicycle dangerous riding behavior recognition result.

2. The method for detecting dangerous riding behaviors of shared bicycles based on image recognition according to claim 1 is characterized in that: The method comprises the following units: A21. Track data acquisition unit: responsible for acquiring the shared bicycle riding track data with timestamps through the shared bicycle positioning system or other data acquisition equipment; A22. Map matching unit: matches the acquired trajectory data to a high-precision map, uses a map matching algorithm to determine the exact location of the trajectory on the map, and improves the relevance and accuracy of the trajectory and the map; A23. Image preprocessing and feature extraction unit: performs cleaning, denoising, and normalization preprocessing operations on the matched trajectory riding image, and extracts key features of the trajectory riding direction in the image; A24. Feature marking unit: marking the extracted trajectory feature parameters as normal riding and dangerous riding trajectories; A25. Dataset production unit: Produce training sets and validation sets according to the feature parameters in proportion, divide the data sets reasonably, and ensure that subsequent model training and validation work can be carried out effectively; A26. Neural network construction and initialization unit: construct a neural network for detecting dangerous riding behaviors of shared bicycles and initialize its network parameters to obtain a primary neural network for detecting dangerous riding behaviors of shared bicycles; A 27. Model training unit: Train the primary shared bicycle dangerous riding behavior detection neural network through the training set, use the optimization algorithm and loss function to continuously adjust the network parameters, and improve the network's ability to learn and identify the characteristics of dangerous riding behavior; A28. Model verification and evaluation unit: put the verification set into the trained shared bicycle dangerous riding behavior detection neural network for verification, and evaluate the network through evaluation indicators, and further optimize the network performance based on the evaluation results; A29. Behavior recognition unit: collects the trajectory data of the shared bicycle to be identified and matches it with the high-precision map to form an image to be identified, preprocesses the image to be identified, and inputs the preprocessed image to be identified into the completed shared bicycle dangerous riding behavior detection neural network to obtain the shared bicycle dangerous riding behavior recognition result.

3. According to claim 1, a method for detecting dangerous riding behaviors of shared bicycles based on image recognition is characterized in that: In step S1, the shared bicycle riding trajectory data with timestamps has timestamps accurate to the second level, and each timestamp is t1, t2, t3...tn with the same interval time, ensuring that the time sequence and duration of the riding can be accurately reflected.

4. The method for detecting dangerous riding behaviors of shared bicycles based on image recognition according to claim 1, characterized in that: In step S2, ensure that the selected high-precision map can accurately present various geographic information of roads, intersections, and landmarks; Through the matching algorithm, each coordinate point in the shared bicycle riding trajectory data and its corresponding timestamp information are accurately matched with the corresponding position on the high-precision map to achieve the accurate "landing point" of the trajectory data on the map.

5. The method for detecting dangerous riding behaviors of shared bicycles based on image recognition according to claim 1, characterized in that: In step S3, the image is normalized to ensure the consistency of image size during subsequent analysis and processing.

6. The method for detecting dangerous riding behaviors of shared bicycles based on image recognition according to claim 1, characterized in that: In step S4, dangerous riding trajectories include riding against traffic, making illegal left turns, and crossing the road.

7. The method for detecting dangerous riding behaviors of shared bicycles based on image recognition according to claim 1, characterized in that: In step S5, the ratio of the training set to the test set is 7:

3.

8. The method for detecting dangerous riding behaviors of shared bicycles based on image recognition according to claim 1, characterized in that: In step S7, the training of the primary shared bicycle dangerous riding behavior detection neural network specifically includes the following steps: S71: normalize all data in the training set; S72: The processed data is put into the neural network and processed by the convolution layer, pooling layer and fully connected layer in sequence; S73: Calculate the loss function of the trained neural network, and adjust the neural network based on the principle of minimizing the loss function.

9. The method for detecting dangerous riding behaviors of shared bicycles based on image recognition according to claim 1, characterized in that: In A28, the formula for the accuracy of the evaluation index for evaluating the network is: , the recall formula is: .

Images