Driver penalty system based on openAI

Abstract

The invention discloses a driver penalty system based on OpenAI, and the driver penalty system based on OpenAI can quickly analyze data (such as journey recording, telephone recording, video data and driving route data) from various sources through an automatic data processing and intelligent analysis module, recognizes illegal behaviors of a driver from the data, and improves the accuracy of the driver penalty. For example, passengers are abused, detour, smoking and the like. The whole process of the system is automatically processed, the links of manual intervention are reduced from data cleaning to illegal behavior judgment and decision generation, and compared with a traditional manual penalty mode, the efficiency is improved, the processing time is shortened, and meanwhile the management and operation cost is reduced.

Description

Technical Field

[0001] The present invention belongs to the field of driver penalty systems, and more specifically, relates to a driver penalty system based on openAI. Background Art

[0002] Driver penalty is to impose different types of penalties (such as fines, order dispatching bans, etc.) for some violations of drivers (such as drivers taking detours, having unpleasant odors in the car, verbally abusing passengers, etc.).

[0003] The AI big data model (artificial intelligence algorithm model), also known as the AI large model, refers to a machine learning model with ultra-large-scale parameters (usually more than 1 billion) and super-strong computing resources. Such a model can process massive amounts of data and complete various complex tasks, such as natural language processing, image recognition, etc. The large model is essentially a deep neural network model trained with massive amounts of data, and its huge data and parameter scale achieve the emergence of intelligence, showing human-like intelligence.

[0004] The current penalty system mainly relies on manual identification and processing of violations. This method not only consumes a large amount of human resources but is also easily affected by subjective factors, making it difficult for the efficiency and accuracy of penalties to meet actual needs.

[0005] In view of this, the present invention is specifically proposed. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and provide a driver penalty system based on openAI, which solves the problems raised in the above background art.

[0007] To solve the above technical problem, the basic concept of the technical solution adopted by the present invention is as follows:

[0008] A driver penalty system based on openAI, comprising: a data acquisition module for collecting driver behavior data through trip recordings, phone recordings, driving route data, and video data;

[0009] A data processing module for normalizing the collected data, including cleaning, annotation, feature extraction, and feature selection;

[0010] An AI analysis module for analyzing the driver's violations using the AI big data model and outputting penalty decisions;

[0011] A penalty execution module for issuing specific penalty measures to the driver according to the output result of the AI analysis module.

[0012] Optionally, the data acquisition module includes a trip recording acquisition unit for acquiring the conversation recording during the trip of the driver and the passenger, a telephone recording acquisition unit for acquiring the telephone recording between the driver and the customer service, a driving route data acquisition unit for obtaining the driving route, staying time and detour behavior of the driver, and a video data acquisition unit for obtaining in-vehicle video information and detecting violations such as smoking and improper grooming.

[0013] Optionally, after the feature extraction is completed, feature selection is required. The steps for selection are as follows:

[0014] Based on the formula evaluate the discrimination ability of the feature, where μ COS and μ CNOS are the means of the violation behavior and the non-violation behavior respectively, and are the variances of the violation behavior and the non-violation behavior respectively;

[0015] Evaluate the discrimination ability of the feature according to the between-class distance and the within-class distance. The between-class distance and the within-class distance are calculated as follows: S B =p COS (μ COS -μ 0 ) 2 +p CNOS (μ CNOS -μ 0 ) 2 , where p COS and p CNOS are the prior probabilities of the violation behavior and the non-violation behavior respectively, and μ 0 is the overall mean;

[0016] Estimate the distribution difference of the feature between the violation behavior and the non-violation behavior through the formula , and then evaluate the significance between categories by calculating the F value of the feature. Its expression is: Finally, sort the features and gradually eliminate the unimportant features, and retain the features that contribute the most to the classification.

[0017] Optionally, the steps for analyzing the driver's violation behavior using the AI big data model and outputting the penalty decision are as follows:

[0018] Fuse the features from different modalities to form a unified input format, and use CNN to analyze audio and image features;

[0019] The time series analysis model is used to analyze the dynamic change model of the driving route data to classify the input data, output the category of the violation behavior, and generate the confidence level or severity score of each violation behavior;

[0020] The output result of the CNN and the output structure of the time series analysis model are adaptively weighted and fused to obtain output features, and the output features are classified to obtain the violation result.

[0021] Optionally, the AI big data model is a deep learning model based on a convolutional neural network. The model architecture includes a convolutional layer, a pooling layer, an activation function layer, and a Softmax layer. Among them, the convolutional layer is used to extract high-dimensional features of the input data; the ReLU activation function is used to introduce non-linearity to enhance the model's ability to learn complex data. Its formula is as follows: a = max(0, η), where η is the input value. The pooling layer reduces the data dimension through local maximum operations and reduces the computational complexity. The Softmax layer is used to generate the probability distribution of each category and selects the category with the highest probability as the model output.

[0022] Optionally, the steps for the time series analysis model to analyze the dynamic change model of driving route data to classify the input data, output the category of violation behavior, and simultaneously generate the confidence level or severity score of each violation behavior are as follows:

[0023] The original data is constructed into a complete time series, and at the same time, dynamic features such as speed change, stay time, and route trajectory deviation are extracted. Then, various features are extracted from the time series, including dynamic features (speed change, acceleration, route deviation), local features (average speed and route change amplitude within the time window), and global statistical features (mean, variance, skewness, and kurtosis);

[0024] The driving route data is input into the LSTM model. The LSTM model analyzes the temporal changes of driving behavior by capturing the dynamic relationships in the time series to identify potential violation behavior patterns;

[0025] Classify the newly input driving data, output the category of driving behavior (such as normal driving, detouring behavior, abnormal parking, etc.), and simultaneously generate the confidence level or severity score of each behavior;

[0026] According to the model classification result, output the category of driving behavior and the corresponding confidence level and severity score, and store the analysis result in the system database.

[0027] Optionally, the LSTM model structure includes an input layer, an LSTM hidden layer, and a fully connected output layer; the hidden layer is used to capture the long-term dependence relationships in the time series data, and the output layer is used to predict the target category or generate regression values.

[0028] Optionally, the steps for issuing specific penalty measures to the driver according to the output result of the AI analysis module are as follows:

[0029] The AI analysis module outputs the types of drivers' violation behaviors and their confidence levels or severity scores. The penalty execution module identifies the specific violation behaviors of the driver (such as verbal abuse, taking a detour, smoking, etc.) based on the output results, and ranks the severity of multiple behaviors;

[0030] The identified violation behaviors are matched with a preset penalty rule library. Each behavior corresponds to specific penalty rules (such as fines, restricting order dispatching, etc.). If the driver is involved in multiple violation behaviors, the penalty execution module will generate a comprehensive penalty plan;

[0031] According to the matched penalty rules, the penalty execution module generates specific penalty measures. For example, warnings, fines, restricting order dispatching, or suspending services, etc. The system also generates a detailed penalty notice text, which includes descriptions of violation behaviors, penalty bases, penalty measures, and appeal methods.

[0032] After adopting the above technical solutions, 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 the advantages described below at the same time:

[0033] 1. The driver penalty system based on OpenAI can quickly analyze data from multiple sources (such as trip recordings, phone recordings, video data, and driving route data) through automated data processing and intelligent analysis modules, and identify the driver's violation behaviors, such as verbal abuse of passengers, taking a detour, smoking, etc. The entire process of the system is automated, from data cleaning to the determination of violation behaviors and the generation of decisions, reducing the links of manual intervention. Compared with the traditional manual penalty method, it not only improves efficiency, but also shortens the processing time, and at the same time reduces management and operation costs.

[0034] 2. By quantitatively evaluating the discrimination ability of features (such as FDR, between-class distance and within-class distance, divergence analysis, and F-value calculation), it is possible to effectively screen out the features that contribute the most to classification. This method not only improves the model's discrimination ability for violation behaviors, but also reduces the data dimension, simplifies the model complexity, avoids the interference of noise data, thereby improving the accuracy of classification and the generalization ability of the model, and at the same time optimizing the calculation efficiency.

[0035] The following further describes in detail the specific implementation manners of the present invention with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The following drawings in the description are only some embodiments. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts. In the accompanying

[0037] FIGURES:

[0038] Figure 1It is a block diagram of a driver penalty system.

[0039] It should be noted that these drawings and textual descriptions are not intended to limit the scope of the concept of the present invention in any way, but to illustrate the concept of the present invention to those skilled in the art by reference to specific embodiments. Detailed implementation manners

[0040] The present invention will now be further described in detail with reference to the accompanying drawings.

[0041] Please refer to Figure 1 As shown, in this embodiment, a driver penalty system based on openAI is provided, including a data acquisition module for acquiring driver behavior data through trip recordings, phone recordings, driving route data, and video data;

[0042] A data processing module for normalizing the acquired data, including cleaning, annotation, feature extraction, and feature selection;

[0043] An AI analysis module for analyzing the driver's violation behaviors using an AI big data model and outputting penalty decisions; by introducing the AI big data model, the present invention can utilize its powerful data processing ability and intelligent analysis ability to deeply mine and analyze multi-dimensional and multi-modal behavior data. For example, it can analyze abusive behaviors in the recording through speech recognition, identify non-standard behaviors such as smoking or improper grooming using video analysis, or detect illegal driving behaviors such as taking a detour and abnormal parking through driving data. The AI big data model can accurately judge the type and degree of violation of each person based on this data and generate fair and efficient penalty decisions. This method can not only reduce manual participation, greatly improve the penalty efficiency, but also ensure the objectivity and consistency of the processing results

[0044] A penalty execution module for issuing specific penalty measures to the driver according to the output result of the AI analysis module.

[0045] In this embodiment, the data acquisition module includes a trip recording acquisition unit for acquiring the conversation recording during the trip between the driver and the passenger, a phone recording acquisition unit for acquiring the phone recording between the driver and the customer service, a driving route data acquisition unit for obtaining the driving path, staying time, and detour behavior of the driver, and a video data acquisition unit for obtaining in-vehicle video information and detecting illegal behaviors such as smoking and improper grooming.

[0046] In this embodiment, after the feature extraction is completed, feature selection needs to be carried out, and the steps during the selection are as follows:

[0047] Based on the formula to evaluate the discrimination ability of the feature, where μ COS and μ CNOSThey are the means of the violation behavior and non - violation behavior respectively. and They are the variances of the violation behavior and non - violation behavior respectively.

[0048] Evaluate the discrimination ability of features according to the between - class distance and within - class distance, where the between - class distance and within - class distance are calculated as follows: S B = p COS (μ COS - μ 0 ) 2 + p CNOS (μ CNOS - μ 0 ) 2 , where, p COS and p CNOS are the prior probabilities of the violation behavior and non - violation behavior respectively, and μ 0 is the overall mean.

[0049] Estimate the distribution difference of features between the violation behavior and non - violation behavior through the formula , and then, evaluate the significance between categories by calculating the F - value of the feature. Its expression is: Finally, sort the features, gradually eliminate unimportant features, and retain the features that contribute the most to classification. Through the quantitative evaluation of the discrimination ability of features (such as FDR, between - class distance and within - class distance, divergence analysis, and F - value calculation), the present invention can effectively screen out the features that contribute the most to classification. This method not only improves the discrimination ability of the model for violation behaviors, but also reduces the data dimension, simplifies the complexity of the model, avoids the interference of noise data, thereby improving the accuracy of classification and the generalization ability of the model, and at the same time optimizing the calculation efficiency.

[0050] In this embodiment, the steps of using the AI big data model to analyze the driver's violation behavior and output a penalty decision are as follows:

[0051] Fuse the features from different modalities to form a unified input format, and use CNN to analyze audio and image features.

[0052] The time series analysis model is used to analyze the dynamic changes of driving route data. The model classifies the input data, outputs the category of violation, and generates a confidence or severity score for each violation. The time series analysis model (such as LSTM) focuses on the dynamic changes of driving route data. By capturing the temporal characteristics of driving behavior (such as speed changes, dwell time, and trajectory deviation), it can effectively identify abnormal driving behaviors, such as detours or abnormal parking. LSTM can mine long-term dependencies in time series and generate accurate behavior classification results and corresponding confidence and severity scores. This method not only covers dynamic violation scenarios in driving behavior, but also provides a clear quantitative basis for subsequent penalties, while improving the system's ability to process dynamic data.

[0053] The output results of CNN and the output structure of the time series analysis model are adaptively weighted to obtain output features, and the output features are classified to obtain violation results. The output results of CNN and the time series analysis model are adaptively weighted to combine static features (such as voice and video) with dynamic features (such as driving route behavior) to give full play to the advantages of the two models. The weighting mechanism automatically adjusts the weights according to the importance of the features, making the fused features more targeted and enhancing the classification model's ability to identify complex violations. This fusion method not only improves the model's ability to process multimodal data, but also makes the system more robust when facing a variety of violation scenarios, significantly improving the reliability and accuracy of classification.

[0054] In this embodiment, the AI ​​big data model is a deep learning model based on a convolutional neural network, and the model architecture includes a convolution layer, a pooling layer, an activation function layer and a Softmax layer, wherein the convolution layer is used to extract the high-dimensional features of the input data; the ReLU activation function is used to introduce nonlinearity to enhance the model's ability to learn complex data, and its formula is as follows: a=max(0,η), wherein η is the input value, the pooling layer reduces the data dimension through the local maximum operation, reduces the computational complexity, and the Softmax layer is used to generate the probability distribution of each category, and selects the category with the largest probability as the model output.

[0055] In this embodiment, the time series analysis model is used to analyze the dynamic change model of the driving route data to classify the input data, output the category of the violation, and generate the confidence or severity score of each violation in the following steps:

[0056] The original data is constructed into a complete time series, and dynamic features such as speed change, dwell time, and route trajectory deviation are extracted at the same time. Then, multiple features are extracted from the time series, including dynamic features (speed change, acceleration, route deviation), local features (average speed and route change amplitude within the time window), and global statistical features (mean, variance, skewness, and kurtosis);

[0057] Input the driving route data into the LSTM model. By capturing the dynamic relationships in the time series, the LSTM model analyzes the temporal changes in driving behavior to identify potential violation patterns.

[0058] Classify the newly input driving data, output the categories of driving behavior (such as normal driving, detouring behavior, abnormal parking, etc.), and simultaneously generate a confidence or severity score for each behavior.

[0059] According to the model classification results, output the driving behavior category and the corresponding confidence and severity scores, and store the analysis results in the system database.

[0060] In this embodiment, the LSTM model structure includes an input layer, an LSTM hidden layer, and a fully connected output layer; the hidden layer is used to capture the long-term dependencies in the time series data, and the output layer is used to predict the target category or generate a regression value.

[0061] In this embodiment, the steps for issuing specific penalty measures to the driver according to the output results of the AI analysis module are as follows:

[0062] The AI analysis module outputs the types of the driver's violation behaviors and their confidence or severity scores. The penalty execution module identifies the specific violation behaviors of the driver (such as verbal abuse, detouring, smoking, etc.) according to the output results, and ranks the severity of multiple behaviors; based on the analysis of audio and image features by CNN, the violation information in the driver's behavior can be fully mined. For example, by analyzing the tone changes or keywords in the audio, the behavior of verbally abusing passengers can be detected; by analyzing the image data, violation behaviors such as the driver smoking or having an untidy appearance in the car can be identified. The convolutional layer of CNN extracts high-dimensional features, the pooling layer reduces the computational complexity, and the Softmax layer outputs accurate classification results. The fusion of different modality data further enhances the comprehensive expression ability of features, enabling the system to accurately identify the driver's behavior from multiple perspectives and providing comprehensive and reliable input for subsequent penalties.

[0063] Match the identified violation behaviors with the preset penalty rule library. Each behavior corresponds to specific penalty rules (such as fines, order assignment restrictions, etc.). If the driver is involved in multiple violation behaviors, the penalty execution module will generate a comprehensive penalty plan.

[0064] According to the matching penalty rules, the penalty execution module generates specific penalty measures. For example, warnings, fines, order dispatching restrictions, or service suspensions, etc. The system also generates a detailed penalty notice text, which includes a description of the violation, the basis for the penalty, the penalty measures, and the appeal method. The AI analysis module outputs the driver's violation behavior and its severity score, and the penalty execution module automatically generates a specific penalty plan in combination with the penalty rule library, such as measures like warnings, fines, order dispatching restrictions, or service suspensions. The system can comprehensively handle multiple violations, generate a reasonable comprehensive penalty plan through severity ranking and rule matching, and at the same time automatically generate a detailed notice text, including the violation description, the basis for the penalty, and the appeal method. This mechanism significantly improves the efficiency and accuracy of penalties, while ensuring the transparency and fairness of the penalty process, and provides clear regulatory guidelines for drivers.

[0065] The present invention is not limited to the above embodiments. Anyone should be aware that structural changes made under the inspiration of the present invention, as long as they have the same or similar technical solutions as the present invention, fall within the protection scope of the present invention. The technologies, shapes, and structures not described in detail in the present invention are all well-known technologies.

Claims

1. A driver penalty system based on openAI, characterized in that: include: Data collection module, used to collect driver behavior data through trip recordings, phone recordings, driving route data and video data; The data processing module is used to standardize the collected data, including cleaning, labeling, feature extraction and feature selection; AI analysis module, which uses AI big data models to analyze drivers’ violations and output penalty decisions; The penalty execution module is used to issue specific penalty measures to the driver based on the output results of the AI ​​analysis module.

2. According to claim 1, a driver penalty system based on openAI is characterized in that: The data collection module includes a travel recording collection unit for collecting recordings of conversations between the driver and passengers during the journey, a telephone recording collection unit for collecting recordings of telephone calls between the driver and customer service, a travel route data collection unit for obtaining the driver's driving path, stop time and detour behavior, and a video data collection unit for obtaining in-vehicle video information and detecting smoking and untidy appearance violations.

3. The driver penalty system based on openAI according to claim 1 is characterized in that: After feature extraction is completed, feature selection is required. The steps for selection are: Based on the formula Evaluate the distinguishing ability of features, where μ COS and μ CNOS are the means of violations and non-violations, respectively. and are the variances of violations and non-violations, respectively; The distinguishing ability of features is evaluated according to the inter-class distance and intra-class distance, where the inter-class distance and intra-class distance are calculated as follows: S B =p COS (μ COS -μ0) 2 +p CNOS (μ CNOS -μ0) 2 , Among them, p COS and p CNOS are the prior probabilities of violation and non-violation, respectively, and μ0 is the population mean; By formula Estimate the distribution difference of the feature between violations and non-violations, and then evaluate the significance between categories by calculating the F value of the feature, which is expressed as: Finally, the features are sorted to gradually eliminate unimportant features and retain the features that contribute most to the classification.

4. The driver penalty system based on openAI according to claim 1 is characterized in that: The steps to use AI big data models to analyze drivers’ violations and output penalty decisions are as follows: The features from different modalities are combined into a unified input format, and CNN is used to analyze audio and image features. Time series analysis models are used to analyze the dynamic changes of driving route data. The model classifies the input data and outputs the category of violation behavior, while generating a confidence or severity score for each violation; The output results of CNN and the output structure of the time series analysis model are adaptively weighted and fused to obtain output features, and the output features are classified to obtain violation results.

5. The driver penalty system based on openAI according to claim 1 is characterized in that: The AI ​​big data model is a deep learning model based on convolutional neural networks. The model architecture includes convolutional layers, pooling layers, activation function layers and Softmax layers. The convolutional layers are used to extract high-dimensional features of the input data. The ReLU activation function is used to introduce nonlinearity to enhance the model's ability to learn complex data, which is expressed as: a=max(0,η), where η is the input value. The pooling layer reduces the data dimension and reduces the computational complexity through local maximum operations. The Softmax layer is used to generate probability distributions for various categories and select the category with the largest probability as the model output.

6. The driver penalty system based on openAI according to claim 1 is characterized in that: The time series analysis model is used to analyze the dynamic changes of driving route data. The model classifies the input data and outputs the category of violation. The steps to generate the confidence or severity score of each violation are: The original data is constructed into a complete time series, and dynamic features such as speed change, dwell time, and route trajectory deviation are extracted at the same time. Then, multiple features are extracted from the time series, including dynamic features, local features, and global statistical features; The driving route data is input into the LSTM model. The LSTM model analyzes the temporal changes of driving behavior by capturing the dynamic relationship in the time series to identify potential violation patterns. Classify the newly input driving data and output the category of driving behavior, while generating a confidence or severity score for each behavior; According to the model classification results, the driving behavior category and the corresponding confidence and severity scores are output, and the analysis results are stored in the system database.

7. The driver penalty system based on openAI according to claim 1 is characterized in that: The LSTM model structure includes an input layer, an LSTM hidden layer, and a fully connected output layer; the hidden layer is used to capture long-term dependencies in time series data, and the output layer is used to predict target categories or generate regression values.

8. The driver penalty system based on openAI according to claim 1 is characterized in that: The steps for issuing specific penalty measures to drivers based on the output results of the AI ​​analysis module are as follows: The AI ​​analysis module outputs the driver's violation type and its confidence or severity score. The penalty execution module identifies the driver's specific violation based on the output results and ranks the severity of the various behaviors. The identified violations are matched with the preset penalty rule library. Each behavior corresponds to a specific penalty rule. If the driver is involved in multiple violations, the penalty execution module will generate a comprehensive penalty plan. Finally, based on the matched penalty rules, the penalty execution module generates specific penalty measures.

Images