Dangerous chemical vehicle identification method considering multi-modal time sequence information

Abstract

The invention belongs to the technical field of hazardous chemical substance vehicle identification, and discloses a hazardous chemical substance vehicle identification method considering multi-modal time sequence information, and the method comprises the following steps: S1, data collection; s2, data preprocessing; s3, feature extraction; s4, modeling a time sequence; s5, multi-modal information fusion is carried out; s6, classifying and identifying; s7, performing real-time feedback and early warning; and S8, updating and optimizing the model. According to the dangerous chemical vehicle identification method considering the multi-modal time sequence information, multi-modal data can be effectively integrated, and the accuracy and real-time performance of dangerous chemical vehicle identification are improved; the false alarm rate and the missing report rate can be reduced, and reliable decision support is provided for a traffic management department; the method can adapt to constantly changing traffic conditions, and high accuracy is kept through real-time updating and optimization of the model; the efficiency and effect of traffic safety management can be remarkably improved, and risks and potential hazards in the dangerous chemical transportation process are reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of hazardous chemical vehicle identification, and particularly relates to a hazardous chemical vehicle identification method considering multimodal time series information. Background Art

[0002] With the rapid development of road transportation, the safety and regulation of vehicles transporting hazardous chemicals are becoming increasingly important. Existing technologies often rely on single-source video surveillance or GPS signals for vehicle identification and tracking, which presents significant limitations. Video surveillance is easily affected by ambient light, resulting in reduced recognition accuracy at night or in inclement weather. GPS signals, on the other hand, are insensitive to micro-dynamic changes in vehicles, making it difficult to accurately predict their behavior and meeting real-time requirements.

[0003] Therefore, there is an urgent need for a new identification method that can accurately and quickly identify hazardous chemical transport vehicles in complex traffic environments to improve the efficiency and accuracy of road safety management. Summary of the Invention

[0004] In light of this, the present invention aims to provide a method for identifying hazardous chemical vehicles that considers multimodal time series information. This method aims to address the low accuracy and poor real-time performance of existing hazardous chemical vehicle identification techniques. This method utilizes multi-source data, including video surveillance information, GPS positioning information, and vehicle behavior patterns, combined with time series analysis techniques, to improve the accuracy of hazardous chemical vehicle identification in various traffic environments, enabling efficient and precise monitoring of hazardous chemical transport vehicles.

[0005] Application scenarios of the present invention include but are not limited to:

[0006] Traffic monitoring environment: Identification of hazardous chemical vehicles under video surveillance restricted conditions such as nighttime or inclement weather;

[0007] Real-time traffic management: Realize dynamic real-time monitoring of hazardous chemical vehicles on highways and other roads;

[0008] Safety early warning system: By analyzing the behavior patterns of hazardous chemical vehicles, timely safety early warning information is issued;

[0009] Transport scheduling optimization: Optimize the driving routes of hazardous chemical transport vehicles to reduce potential risks.

[0010] The present invention provides a method for identifying hazardous chemical vehicles considering multimodal time series information, comprising the following steps:

[0011] S1. Data collection: Real-time collection of multimodal data from hazardous chemical vehicles;

[0012] S2. Data preprocessing: Preprocess the data collected in step S1 to improve data quality;

[0013] S3. Feature extraction: Extract vehicle features, trajectory features, and environmental features based on the preprocessed data in step S2;

[0014] S4. Time Series Modeling: Based on the features extracted in step S3, a time series model is constructed using a long short-term memory (LSTM) network or a gated recurrent unit (GRU). This time series model is used to learn the movement patterns and behavior patterns of hazardous chemical vehicles.

[0015] S5. Multimodal information fusion: Based on the time series information of the hazardous chemical vehicle obtained in step S4, the attention mechanism or multimodal fusion network technology is used to fuse the time series information of the hazardous chemical vehicle's vision, trajectory, and environment to obtain a comprehensive feature vector;

[0016] S6. Classification and Identification: Based on the comprehensive feature vector obtained in step S5, a deep learning algorithm is used to classify and identify hazardous chemical vehicles and other vehicles.

[0017] S7. Real-time feedback and warning: The recognition results of step S6 are fed back to the real-time traffic management system, triggering a safety warning mechanism to help traffic management departments take timely measures;

[0018] S8. Model update and optimization: Through continuous learning mechanisms and online learning algorithms, the model is continuously updated to adapt to changes in the traffic environment and new data patterns.

[0019] Furthermore, the multimodal data of the hazardous chemical vehicle includes video surveillance data, GPS trajectory data, and vehicle status and environmental sensor data of the hazardous chemical vehicle.

[0020] Furthermore, in step S2, the video and image data collected in step S1 are pre-processed, including image enhancement, signal noise reduction, and outlier detection and filling.

[0021] Furthermore, the specific content of step S3 is:

[0022] Extract vehicle features from video surveillance data, including model, color, and license plate;

[0023] Extract trajectory features from GPS data, including speed, acceleration, and steering angle;

[0024] Extract environmental features, including temperature and humidity, from sensor data.

[0025] Furthermore, the deep learning algorithm in step S6 is a convolutional neural network (CNN).

[0026] Beneficial effects:

[0027] This paper proposes a method for reconstructing the longitudinal trajectory of highway vehicles that takes into account instantaneous speed estimation errors. This method effectively integrates multimodal data and improves the accuracy and real-time performance of hazardous chemical vehicle identification. In complex traffic environments, this method can reduce false alarm and missed alarm rates, providing reliable decision-making support for traffic management departments. Furthermore, the method can adapt to changing traffic conditions and maintain high accuracy through real-time model updates and optimization. Ultimately, this method can significantly improve the efficiency and effectiveness of traffic safety management and reduce the risks and potential hazards during hazardous chemical transportation.

[0028] Other advantages, objects, and features of the present invention will be described in part in the following description and, in part, will be apparent to those skilled in the art upon examination of the following description or may be learned from practice of the present invention. The objects and other advantages of the present invention may be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 This is a flow chart of a hazardous chemical vehicle identification method that considers multimodal time series information according to the present invention. DETAILED DESCRIPTION

[0030] To make the technical solutions, advantages, and purposes of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the described embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0031] like Figure 1 As shown, the present invention provides a method for identifying hazardous chemical vehicles considering multimodal time series information, comprising the following steps:

[0032] S1. Data collection: Real-time collection of multimodal data from hazardous chemical vehicles;

[0033] Multimodal data of hazardous chemical vehicles includes video surveillance data, GPS trajectory data, and vehicle status and environmental sensor data;

[0034] Multiple video surveillance cameras and GPS receivers are installed on a highway. The cameras capture image data every five seconds, while the GPS receivers update the vehicle's location, speed, and direction every second. Data from vehicle and environmental sensors is transmitted in real time via a wireless network to a central data processing system.

[0035] S2. Data preprocessing: Preprocess the data collected in step S1 to improve data quality;

[0036] Preprocessing of the video and image data collected in step S1, including image enhancement, signal noise reduction, and outlier detection and filling;

[0037] At night or in inclement weather, images captured by video surveillance cameras may be noisy and blurry. Preprocessing uses image enhancement algorithms to improve image quality and object detection algorithms to identify vehicles in images. GPS data may contain data loss in some cases, and preprocessing uses interpolation to fill in these missing data and ensure data continuity.

[0038] S3. Feature extraction: Extract vehicle features, trajectory features, and environmental features based on the preprocessed data in step S2;

[0039] Use convolutional neural networks (CNNs) to extract vehicle features from video surveillance data, including visual features such as vehicle model, color, and license plate;

[0040] Extract trajectory features from GPS data, including dynamic features such as the vehicle's average speed, acceleration, and steering angle;

[0041] Extract environmental features from sensor data, including the vehicle's driving environment temperature, humidity and other information.

[0042] S4. Time Series Modeling: Based on the features extracted in step S3, a time series model is constructed using a long short-term memory (LSTM) network or a gated recurrent unit (GRU). This time series model is used to learn the movement patterns and behavior patterns of hazardous chemical vehicles and capture the long-term dependencies of their behavior patterns.

[0043] On highways, vehicle movement patterns exhibit distinct time-series characteristics. Time series modeling uses an LSTM network to model vehicle GPS trajectory data, learning the changing patterns of vehicle acceleration and steering angles. Through training, the LSTM network can predict the vehicle's future movement trajectory and behavior patterns.

[0044] S5. Multimodal information fusion: Based on the time series information of the hazardous chemical vehicle obtained in step S4, the attention mechanism or multimodal fusion network technology is used to fuse the time series information of the hazardous chemical vehicle's vision, trajectory, and environment to obtain a comprehensive feature vector, thereby achieving deep fusion of multi-source heterogeneous data;

[0045] During the multimodal information fusion process, an attention mechanism is used to dynamically weight features from different modalities. For example, at night or in inclement weather, video surveillance data may be weighted less, while GPS and sensor data may be weighted more. By fusing features from these modalities, a comprehensive feature vector is generated that more accurately describes the status of the hazardous chemical vehicle.

[0046] S6. Classification and Identification: Based on the comprehensive feature vector obtained by fusion in step S5, a deep learning algorithm (such as a convolutional neural network (CNN)) is used to classify and identify hazardous chemical vehicles from other vehicles.

[0047] On highways, a CNN is used to classify and identify the fused feature vectors. Through training, the CNN can distinguish between hazardous chemical vehicles and other ordinary vehicles. The output of step S6, the classification and identification, includes the vehicle category (hazardous chemical vehicle or ordinary vehicle) and a confidence score.

[0048] S7. Real-time feedback and warning: The recognition results of step S6 are fed back to the real-time traffic management system, triggering a safety warning mechanism to help traffic management departments take timely measures;

[0049] When a hazardous chemical vehicle is detected on a highway through classification and identification, an early warning message is immediately sent to the traffic management department. The traffic management department can then view the location and driving trajectory of the hazardous chemical vehicle through the real-time traffic management system and take appropriate safety measures.

[0050] S8. Model Update and Optimization: Through continuous learning mechanisms and online learning algorithms, the model is continuously updated to adapt to changes in the traffic environment and new data patterns;

[0051] On highways, new data is regularly collected and used to retrain and continuously update the LSTM and CNN models. Through online learning algorithms, the models can quickly adapt to new traffic conditions and data patterns, improving the accuracy and real-time performance of hazardous chemical vehicle identification.

[0052] It is hereby stated that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the purpose and scope of the technical solutions, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for identifying hazardous chemical vehicles considering multimodal time series information, characterized in that: The following steps are involved: S1. Data collection: real-time collection of multi-modal data of hazardous chemicals vehicles; S2. Data preprocessing: preprocess the data collected in step S1 to improve data quality; S3. Feature extraction: extracting vehicle features, trajectory features and environmental features based on the preprocessed data in step S2; S4. Time series modeling: Based on the features extracted in step S3, a time series model is constructed using a long short-term memory network (LSTM) or a gated recurrent unit (GRU). The time series model is used to learn the movement patterns and behavior patterns of hazardous chemical vehicles. S5. Multimodal information fusion: Based on the time series information of the hazardous chemicals vehicle obtained in step S4, the time series information of the hazardous chemicals vehicle vision, trajectory, and environment is fused using an attention mechanism or a multimodal fusion network technology to obtain a comprehensive feature vector; S6. Classification and identification: Based on the comprehensive feature vector obtained by fusion in step S5, the hazardous chemical vehicles and other vehicles are classified and identified through the deep learning algorithm; S7. Real-time feedback and warning: Feedback the recognition results of step S6 to the real-time traffic management system, triggering the safety warning mechanism to help the traffic management department take timely measures; S8. Model update and optimization: Through continuous learning mechanisms and online learning algorithms, the model is continuously updated to adapt to changes in the traffic environment and new data patterns.

2. According to claim 1, a method for identifying hazardous chemical vehicles taking into account multimodal time series information is characterized in that: The multimodal data of the hazardous chemicals vehicle includes video surveillance data, GPS trajectory data, and vehicle status and environmental sensor data of the hazardous chemicals vehicle.

3. According to claim 2, a method for identifying hazardous chemical vehicles taking into account multimodal time series information is characterized in that: In step S2, the video and image data collected in step S1 are preprocessed, including image enhancement, signal noise reduction, and outlier detection and filling.

4. According to claim 3, a method for identifying hazardous chemical vehicles taking into account multimodal time series information is characterized in that: The specific content of step S3 is: Extract vehicle features from video surveillance data, including model, color, and license plate; Extract trajectory features from GPS data, including speed, acceleration, and steering angle; Extract environmental features including temperature and humidity from sensor data.

5. According to claim 2, a method for identifying hazardous chemical vehicles taking into account multimodal time series information is characterized in that: The deep learning algorithm in step S6 is a convolutional neural network (CNN).

Images