Automobile lease management method and system based on visual monitoring

Abstract

The invention relates to the technical field of vehicle rental operation management, and discloses a vehicle rental management method and system based on visual monitoring, and the method comprises the steps: S1, multi-modal data collection: obtaining the order information of a vehicle, the position data of the vehicle, and the visual monitoring data inside and outside the vehicle in real time; s2, dynamic risk assessment: constructing a risk scoring model based on the collected multi-modal data, and generating a real-time risk index; the reference indexes of the risk scoring model comprise order risks, vehicle body states and driving behaviors; s3, lease strategy dynamic adjustment: lease terms are adjusted according to the real-time risk index, and the adjustment content comprises the deposit amount and the order distribution logic. The system corresponds to the method. According to the invention, intelligent management is realized through the multi-modal data acquisition module, the dynamic risk assessment module and the leasing strategy dynamic adjustment module, and it is ensured that a user accurately identifies potential risks when leasing a vehicle and the use safety of the vehicle after the vehicle is leased.

Description

Technical Field

[0001] The present application relates to the technical field of vehicle rental operation management, and specifically to a vehicle rental management method and system based on visual monitoring. Background Art

[0002] Traditional car rental management primarily relies on a single data source for basic management functions. Typical solutions include the EasyCVR system and manual management. The EasyCVR system only tracks vehicles through location data or video surveillance, lacking comprehensive analysis of order information, vehicle status, and driving behavior, and unable to build a multi-dimensional risk assessment system. Manual management solutions, on the other hand, rely on managers to assess the risk of renters based on driver's licenses and subjective judgment. These solutions are unable to determine rental risk based on a renter's rental history or actual driving behavior, and lack the ability to predict and proactively intervene in potential risks during the rental process.

[0003] Based on this, it can be seen that the existing car rental management relies only on positioning or video data, resulting in a single risk assessment dimension and difficulty in discovering potential risk associations.

[0004] Therefore, there is an urgent need for an intelligent car rental management technology. Summary of the Invention

[0005] The purpose of this application is to provide a car rental management method and system based on visual monitoring to solve the technical problems raised in the above background technology.

[0006] To achieve the above objectives, this application discloses the following technical solutions:

[0007] In a first aspect, this embodiment discloses a car rental management method based on visual monitoring, the method comprising the following steps:

[0008] S1-Multimodal Data Collection: Real-time acquisition of vehicle order information, vehicle location data, and visual monitoring data inside and outside the vehicle. Order information includes rental time, location, user credit score, and historical violation records. Visual monitoring data includes driver behavior data and vehicle damage data.

[0009] S2-Dynamic Risk Assessment: Build a risk scoring model based on collected multimodal data to generate a real-time risk index. The reference indicators of the risk scoring model include: order risk, vehicle status, and driving behavior;

[0010] S3-Dynamic adjustment of leasing strategy: Adjust the leasing terms based on the real-time risk index, including the deposit amount and order allocation logic.

[0011] Preferably, in S1:

[0012] The acquisition of driver behavior data includes: analyzing visible light camera images based on the YOLOv8 model to detect drunk driving keywords in the driver's posture and voice call content;

[0013] The acquisition of the vehicle body damage data includes: when the vehicle is delivered, combining the YOLOv8 model and lidar point cloud data to reconstruct the vehicle body 3D model to detect body scratches and tire wear, and verify the health of the vehicle body through OBD data.

[0014] Preferably, the acquisition of vehicle body damage data further includes damage classification, and the damage classification specifically includes:

[0015] The degree of damage is divided according to the detection results of body scratches and tire wear, and the repair priority is divided based on the degree of damage.

[0016] Preferably, the risk scoring model is optimized by:

[0017] Multi-dimensional data fusion: Integrate order information, vehicle location data, and visual monitoring data inside and outside the vehicle to build a real-time risk index;

[0018] Machine learning model: Uses the Transformer network to perform time series analysis on driving behavior and combines it with the LSTM network to predict driving risks;

[0019] Closed-loop feedback mechanism: risk assessment results are fed back into the order allocation logic, prioritizing matching low-risk users with low-risk vehicles.

[0020] Preferably, the real-time risk index is calculated by the following formula:

[0021]

[0022] Among them: RI is the real-time risk index, Softmax is the normalized exponential function, δ i is the sensitivity coefficient of environmental risk factors, E i is the i-th environmental risk factor, OR, CR and DR are the standardized scores of order risk, vehicle condition and driving behavior respectively, and w1, w2 and w3 are the trainable weight parameters of the model.

[0023] Preferably, the training process of the risk scoring model includes:

[0024] S21-Dataset Construction: Build a multi-dimensional training dataset based on historical order data, accident records, and vehicle status;

[0025] Model optimization: Using transfer learning technology to optimize models based on public datasets;

[0026] Real-time update: New data is learned online through edge computing devices to continuously optimize the model parameters of the risk scoring model.

[0027] Preferably, the updating of the model parameters is specifically as follows:

[0028]

[0029] Among them, θ t is the model parameter at time t, η is the learning rate, β is the weight of historical data, m is the total number of samples, Attn i is the attention weight of the i-th sample, Confidence i is the sample confidence, is the gradient vector of a single sample.

[0030] Preferably, the adjustment of the lease terms specifically includes:

[0031] Deposit floating mechanism: Based on the results of the real-time risk index, users with a real-time risk index less than 30 will be provided with a deposit reduction policy, users with a real-time risk index between 30-70 will be provided with a basic deposit policy, and users with a real-time risk index greater than 70 will be provided with a double deposit policy;

[0032] Order allocation optimization: users with a real-time risk index greater than 50 are given the right to rent low-value vehicles, and the rental period can be set; users with a real-time risk index greater than 50 are given priority allocation rights for high-value vehicles, and the rental period is not limited.

[0033] Preferably, the S3 further includes an abnormal emergency handling mechanism, which specifically includes:

[0034] When abnormal vehicle start-up, external impact, or deviation from the preset route is detected, a multi-level alarm strategy is triggered. The multi-level alarm strategy includes sending an alarm SMS to the rental platform and pushing reminders to the user's APP, and freezing the rental rights until manual review.

[0035] In a second aspect, the present application discloses a car rental management system based on visual monitoring, which applies the car rental management method based on visual monitoring as described above. The system includes:

[0036] The multimodal data acquisition module includes an order information unit, a positioning unit, and a visual monitoring unit. The order information unit is configured to obtain the rental time, location, user credit score, and historical violation records. The positioning unit is configured to obtain the vehicle's real-time location through a high-precision GPS or Beidou positioning chip. The visual monitoring unit includes an in-vehicle visible light camera, an in-vehicle infrared camera, a wide-angle camera on the vehicle body, and a laser radar installed at a detection point, and is configured to obtain driver behavior data and vehicle damage data.

[0037] The dynamic risk assessment module is configured to: construct a risk scoring model based on the collected multimodal data to generate a real-time risk index; the reference indicators of the risk scoring model include: order risk, vehicle status and driving behavior;

[0038] The leasing strategy dynamic adjustment module is configured to adjust the leasing terms according to the real-time risk index, and the adjustment content includes the deposit amount and order allocation logic.

[0039] Beneficial effects: The car rental management method and system based on visual monitoring of the present application collect multimodal data formed by order information, vehicle location data, and visual monitoring data inside and outside the vehicle, and realize comprehensive analysis of rental business data, spatial location data, and vehicle status data, avoiding the limitations of a single data source. In addition, a risk scoring model is constructed based on multimodal data and a real-time risk index is generated to realize risk assessment of users' rental vehicles and their use, thereby improving vehicle use safety and vehicle safety protection. At the same time, combined with the dynamic adjustment of rental strategies, targeted strategy adjustments are made to rental users with potential risks, thereby improving vehicle safety protection and improving car rental management efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without paying any creative work.

[0041] Figure 1 This is a flowchart of the car rental management method based on visual monitoring provided in an embodiment of the present application. DETAILED DESCRIPTION

[0042] The following is a clear and complete description of the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments of this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0043] In this document, the term "comprising" is intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising..." does not preclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0044] In the first aspect, this embodiment discloses a method such as Figure 1 The car rental management method based on visual monitoring is designed to ensure accurate identification of potential risks when users rent a vehicle and the safety of the vehicle after it is rented (including timely discovery of vehicle damage and vehicle driving safety). Specifically, the method includes the following steps:

[0045] S1-Multimodal Data Collection: Real-time acquisition of vehicle order information, vehicle location data, and visual monitoring data inside and outside the vehicle. Order information includes rental time, location, user credit score, and historical violation records. Visual monitoring data includes driver behavior data and vehicle damage data.

[0046] S2-Dynamic Risk Assessment: Build a risk scoring model based on collected multimodal data to generate a real-time risk index. The risk scoring model uses reference indicators including order risk, vehicle status, and driving behavior. The real-time risk index ranges from 0 to 100 and can be divided into low risk (0-30 points), medium risk (31-70 points), and high risk (71-100 points).

[0047] S3-Dynamic adjustment of leasing strategy: Adjust the leasing terms based on the real-time risk index, including the deposit amount and order allocation logic.

[0048] As described above, the visual monitoring-based car rental management method of this embodiment achieves intelligent management through three modules: multimodal data collection, dynamic risk assessment, and dynamic adjustment of rental strategies. Multimodal data, comprising order information, vehicle location data, and visual monitoring data from both inside and outside the vehicle, is collected to enable comprehensive analysis of rental business data, spatial location data, and vehicle status data, avoiding the limitations of a single data source. Furthermore, a risk scoring model is constructed based on this multimodal data, generating a real-time risk index. This allows for risk assessment of users' vehicle rentals and usage, improving vehicle safety and security. Furthermore, combined with dynamic adjustment of rental strategies, targeted policy adjustments are implemented for potentially risky users, further enhancing vehicle safety and improving car rental management efficiency. Risk assessment results directly drive adjustments to rental terms, forming a closed loop of data, assessment, and decision-making.

[0049] In one embodiment, the acquisition of the driver behavior data includes: analyzing visible light camera images based on the YOLOv8 model, detecting drunk driving keywords in the driver's posture and voice call content; among them, using the YOLOv8 model to analyze visible light camera images, detecting postures such as fatigue (such as closing eyes, yawning) and distraction (such as deviation of vision), and combining voice recognition technology to extract drunk driving keywords (such as "drinking" and "driving").

[0050] Specifically, the driving safety risk index corresponding to the driver behavior data is calculated using the following formula:

[0051]

[0052] Among them, FRI is the driving safety risk index, which ranges from 0 to 1. i To analyze the degree to which the driver's line of sight deviates from the center area in the i-th frame image through YOLOv8 image analysis, the Euclidean distance between the line of sight and the center of the screen is normalized and the value is 0-1. The greater the deviation, the higher the value; t i -t0 is the time difference between the current frame and the first detection of fatigue signal (such as the first yawn), λ is the fatigue attenuation factor, which is obtained by fitting the historical fatigue data (such as the annotated fatigue driving video) using the least squares method. is the time attenuation factor of the fatigue signal, the recently detected gaze deviation Gaze i The weight is higher to simulate the cumulative effect of fatigue state. For example, the contribution of continuous vision deviation within 10 minutes to FRI is significantly higher than that of a single deviation 1 hour ago. j Detect the duration of yawning for video frames, extracted by OpenCV video stream analysis; Intensity j yawn intensity rating, ranging from 1 to 5, is calculated from audio decibels, addressing the one-sidedness of traditional methods that rely solely on duration. BlinkThreshold(t) is a dynamic blink threshold that increases linearly with driving time (e.g., an initial threshold of 15 blinks / minute increases by 2 blinks / hour), adapting to changes in human circadian rhythms. This calculation method improves fatigue monitoring efficiency and quantifies fatigue risk.

[0053] The acquisition of vehicle damage data includes: combining the YOLOv8 model and LiDAR point cloud data to reconstruct a 3D model of the vehicle body at the time of vehicle delivery to detect scratches and tire wear, and verifying the vehicle's health through OBD data. This involves combining YOLOv8 image recognition with LiDAR point cloud data to reconstruct a 3D model of the vehicle body, detect scratches, dents, tire wear and other damage, and verify the detection results (such as abnormal tire pressure caused by tire wear) through OBD data (tire pressure, engine status).

[0054] Furthermore, the acquisition of vehicle body damage data also includes damage classification, and the damage classification specifically includes:

[0055] The damage degree is divided according to the detection results of vehicle body scratches and tire wear, and the repair priority is divided based on the damage degree. According to the damage degree, it is divided into minor damage (such as scratches) corresponding to DS<3, moderate damage (such as tire wear) corresponding to 3≤DS<7, and severe damage (such as structural deformation) corresponding to DS≥7. Different repair priorities are matched, such as forced return to the factory for severe damage.

[0056] It is feasible to calculate the comprehensive vehicle body damage score by the following formula:

[0057]

[0058] DS is the damage score, which ranges from 0 to 10, and Volume k is the three-dimensional volume of the kth damage calculated by the differential integral of the LiDAR point cloud and the original model. Compared with the traditional two-dimensional area measurement, it can better reflect the damage depth (for example, the damage volume of a dent with a depth of 5 mm is significantly larger than that of a surface scratch); TotalVolume is the standard volume of the vehicle body, Severity k is the severity coefficient of the damage type (e.g. scratch is 1, deformation is 3, structural damage is 5), OBD anomaly The OBD data anomaly is calculated by calculating the Z-score of the OBD parameters and the historical mean, with a value ranging from 0 to 1, and is calculated using Gaussian distribution outlier detection. TimeFactor is the damage duration factor (e.g., a value of 1.5 for new damage emphasizes immediate damage liability determination; a value of 1.0 for old damage avoids repeated assessments if the damage existed before the lease). This calculation method improves the accuracy of vehicle damage detection and reduces false positives through OBD reading and re-verification. It also enables automatic damage classification, thereby improving repair and accountability efficiency and reducing manual review costs.

[0059] In one embodiment, the risk scoring model is optimized by:

[0060] Multi-dimensional data fusion: Integrates order information (e.g., determining user creditworthiness), vehicle location data (e.g., determining route deviation / exceeding permitted areas), and visual monitoring data inside and outside the vehicle (driving behavior + vehicle damage) to construct a real-time risk index.

[0061] Machine learning model: Uses a Transformer network to perform time series analysis on driving behavior (such as the time series of continuous fatigue signals) and combines it with an LSTM network to predict driving risks.

[0062] Closed-loop feedback mechanism: The risk assessment results are reversely input into the order allocation logic to prioritize matching low-risk users with low-risk vehicles. For example, high-risk users are prioritized with low-value vehicles (such as vehicles older than 5 years) to reduce the risk of asset loss.

[0063] Furthermore, the real-time risk index is calculated using the following formula:

[0064]

[0065] Where: RI is the real-time risk index (0-100), Softmax is the normalized exponential function, which realizes the multi-factor competition weight, performs normalized exponential weighting on OR / CR / DR, and highlights the dominant risk factor. For example, when DR = 90 (serious driving behavior risk), its weight w3 is automatically increased to more than 60%, suppressing the influence of other factors; δ i is the sensitivity coefficient of environmental risk factors, obtained through regression of historical accident data, E i The real-time data is obtained from the meteorological API (such as OpenWeatherMap) and the traffic big data platform, which is the i-th environmental risk factor (with a value of 0-1, such as heavy rain = 0.8, construction section = 0.6), the environmental factor E i For example, the sensitivity coefficient δ of heavy rain and construction sections can be obtained by regression of historical accident data. i , for example, the rainstorm factor E i =0.8 and δ i = 1.2 increases the RI by an additional 96%, simulating the amplifying effect of extreme environments on risk. OR, CR, and DR are standardized scores for order risk, vehicle condition, and driving behavior, respectively. Specifically, OR is discrete features such as the user's credit score and number of past violations converted to continuous values ​​using WOE encoding; CR is the DS score normalized to 0-1; and DR is the FRI index mapped to 0-100. w1, w2, and w3 are the model's trainable weight parameters, used to quantify the contribution of order risk, vehicle condition, and driving behavior to the final risk index. These are internal parameters of the machine learning model, stored in the weight matrix of the Transformer or LSTM network and automatically optimized through a data-driven approach, rather than manually preset fixed values. This calculation method improves the efficiency and reliability of risk prediction.

[0066] It is feasible that the training process of the risk scoring model includes:

[0067] S21-Dataset Construction: Build a multi-dimensional training dataset based on historical order data, accident records, and vehicle status. For example, we can integrate over 100,000 rental orders (including user credit and rental duration), over 2,000 accident records (including time, location, and damage extent), and over 50,000 vehicle images (labeled with damage type) to build a cross-modal training set.

[0068] Model optimization: Using transfer learning technology, we optimize the model based on public datasets (such as KITTI) to improve generalization capabilities;

[0069] Real-time update: New data is learned online through edge computing devices to continuously optimize the model parameters of the risk scoring model.

[0070] Furthermore, the updating of the model parameters is specifically as follows:

[0071]

[0072] Among them, θ t is the model parameter at time t, η is the learning rate, β is the historical data weight, η and β are determined by cross-validation to determine the initial value, and then dynamically adjusted by the online annealing algorithm; m is the total number of samples, Attn i is the attention weight of the i-th sample, generated by the Transformer self-attention mechanism, Confidence i The confidence level of the sample ranges from 0 to 1. Based on the cosine similarity between the predicted result and the true label, samples with a confidence level lower than 0.5 (such as blurred images) are filtered out to prevent noise data from contaminating the model. is the gradient vector of a single sample, calculated by the backpropagation algorithm. This allows the attention mechanism to filter high-value samples (such as risk events with RI > 80), prioritize model parameter updates, and prevent old data from overwhelming new trends. This improves the model's response time to emerging risk patterns (such as new distracting behaviors) while ensuring accurate model judgments.

[0073] In one embodiment, the adjustment of the lease terms specifically includes:

[0074] Deposit floating mechanism: Based on the results of the real-time risk index, users with a real-time risk index less than 30 will be provided with a deposit reduction policy, users with a real-time risk index between 30-70 will be provided with a basic deposit policy, and users with a real-time risk index greater than 70 will be provided with a double deposit policy;

[0075] Order allocation optimization: users with a real-time risk index greater than 50 are given the right to rent low-value vehicles, and the rental period can be set; users with a real-time risk index greater than 50 are given priority allocation rights for high-value vehicles, and the rental period is not limited.

[0076] By adjusting the lease terms, we can improve the corresponding treatment of users with potential risks, use deposit strategies to eliminate low-quality users (such as those with tight funds), and reduce the risk of losses and irrecoverable losses.

[0077] In one embodiment, S3 further includes an abnormal emergency handling mechanism, which specifically includes:

[0078] When abnormal vehicle start-up, external impact, or deviation from the preset route is detected, a multi-level alarm strategy is triggered. The multi-level alarm strategy includes sending an alarm SMS to the rental platform and pushing reminders to the user's APP, and freezing the rental rights until manual review.

[0079] In this way, after the vehicle is rented, risk warnings can be issued to managers and safety warnings can be issued to users in the first place, thereby improving the safety of vehicle rental and vehicle use.

[0080] In a second aspect, this embodiment provides a car rental management system based on visual monitoring, which applies the car rental management method based on visual monitoring described above. The system includes:

[0081] The multimodal data acquisition module includes an order information unit, a positioning unit, and a visual monitoring unit. The order information unit is configured to obtain the rental time, location, user credit score, and historical violation records. The positioning unit is configured to obtain the vehicle's real-time location using a high-precision GPS or Beidou positioning chip. The visual monitoring unit includes an in-vehicle visible light camera, an in-vehicle infrared camera, a body-mounted wide-angle camera, and a laser radar installed at a detection point (which can be installed at a gantry at a delivery point, such as a delivery point, to obtain corresponding radar detection data when the vehicle passes through the gantry). It is configured to obtain driver behavior data and body damage data.

[0082] The dynamic risk assessment module is configured to: construct a risk scoring model based on the collected multimodal data to generate a real-time risk index; the reference indicators of the risk scoring model include: order risk, vehicle status and driving behavior;

[0083] The leasing strategy dynamic adjustment module is configured to adjust the leasing terms according to the real-time risk index, and the adjustment content includes the deposit amount and order allocation logic.

[0084] It should be noted that the car rental management system based on visual monitoring of this embodiment corresponds to the aforementioned car rental management method based on visual monitoring. Therefore, the undisclosed parts of the car rental management system based on visual monitoring of this embodiment (including but not limited to specific technical solutions, working principles and technical effects) can be referred to the relevant records in the aforementioned car rental management method based on visual monitoring, and this text will not elaborate on them here.

[0085] In the embodiments provided herein, it should be understood that the embodiments described herein can be implemented in hardware, software, firmware, middleware, code, or any appropriate combination thereof. For hardware implementation, the processor can be implemented in one or more of the following units: an application-specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field programmable gate array (FPGA), a processor, a controller, a microcontroller, a microprocessor, other electronic units designed to implement the functions described herein, or a combination thereof. For software implementation, part or all of the processes of the embodiments can be completed by instructing the relevant hardware through a computer program. When implemented, the above program can be stored in a computer-readable storage medium or transmitted as one or more instructions or codes on a computer-readable storage medium. Computer-readable storage media include computer storage media and communication media, wherein the communication media include any medium that facilitates the transmission of a computer program from one place to another. The storage medium can be any available medium that a computer can access. The computer-readable storage medium can include, but is not limited to, RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage media or other magnetic storage devices, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer.

[0086] Finally, it should be noted that the above is only a preferred embodiment of the present application and is not intended to limit the present application. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or make equivalent replacements for some of the technical features therein. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application.

Claims

1. A car rental management method based on visual monitoring, characterized in that: The method comprises the following steps: S1-Multimodal Data Collection: Real-time acquisition of vehicle order information, vehicle location data, and visual monitoring data inside and outside the vehicle. Order information includes rental time, location, user credit score, and historical violation records. Visual monitoring data includes driver behavior data and vehicle damage data. S2-Dynamic Risk Assessment: Build a risk scoring model based on collected multimodal data to generate a real-time risk index. The reference indicators of the risk scoring model include: order risk, vehicle status, and driving behavior; S3-Dynamic adjustment of leasing strategy: Adjust the leasing terms based on the real-time risk index, including the deposit amount and order allocation logic.

2. The car rental management method based on visual monitoring according to claim 1 is characterized in that: In said S1: The acquisition of driver behavior data includes: analyzing visible light camera images based on the YOLOv8 model to detect drunk driving keywords in the driver's posture and voice call content; The acquisition of the vehicle body damage data includes: when the vehicle is delivered, combining the YOLOv8 model and lidar point cloud data to reconstruct the vehicle body 3D model to detect body scratches and tire wear, and verify the health of the vehicle body through OBD data.

3. The car rental management method based on visual monitoring according to claim 2 is characterized in that: The acquisition of the vehicle body damage data also includes damage classification, and the damage classification specifically includes: The degree of damage is divided according to the detection results of body scratches and tire wear, and the repair priority is divided based on the degree of damage.

4. The car rental management method based on visual monitoring according to claim 1 is characterized in that: The risk scoring model is optimized in the following ways: Multi-dimensional data fusion: Integrate order information, vehicle location data, and visual monitoring data inside and outside the vehicle to build a real-time risk index; Machine learning model: Uses the Transformer network to perform time series analysis on driving behavior and combines it with the LSTM network to predict driving risks; Closed-loop feedback mechanism: risk assessment results are fed back into the order allocation logic, prioritizing matching low-risk users with low-risk vehicles.

5. The car rental management method based on visual monitoring according to claim 4 is characterized in that: The real-time risk index is calculated using the following formula: Among them: RI is the real-time risk index, Softmax is the normalized exponential function, δ i is the sensitivity coefficient of environmental risk factors, E i is the i-th environmental risk factor, OR, CR and DR are the standardized scores of order risk, vehicle condition and driving behavior respectively, and w1, w2 and w3 are the trainable weight parameters of the model.

6. The car rental management method based on visual monitoring according to claim 4 is characterized in that: The training process of the risk scoring model includes: S21-Dataset Construction: Build a multi-dimensional training dataset based on historical order data, accident records, and vehicle status; Model optimization: Using transfer learning technology to optimize models based on public datasets; Real-time update: New data is learned online through edge computing devices to continuously optimize the model parameters of the risk scoring model.

7. The car rental management method based on visual monitoring according to claim 6 is characterized in that: The updating of the model parameters is specifically as follows: Among them, θ t is the model parameter at time t, η is the learning rate, β is the weight of historical data, m is the total number of samples, Attn i is the attention weight of the i-th sample, Confidence i is the sample confidence, is the gradient vector of a single sample.

8. The car rental management method based on visual monitoring according to claim 1, characterized in that: The adjustments to the lease terms specifically include: Deposit floating mechanism: Based on the results of the real-time risk index, users with a real-time risk index less than 30 will be provided with a deposit reduction policy, users with a real-time risk index between 30-70 will be provided with a basic deposit policy, and users with a real-time risk index greater than 70 will be provided with a double deposit policy; Order allocation optimization: users with a real-time risk index greater than 50 are given the right to rent low-value vehicles, and the rental period can be set; users with a real-time risk index greater than 50 are given priority allocation rights for high-value vehicles, and the rental period is not limited.

9. The car rental management method based on visual monitoring according to claim 1, characterized in that: S3 also includes an abnormal emergency handling mechanism, which specifically includes: When abnormal vehicle start-up, external impact, or deviation from the preset route is detected, a multi-level alarm strategy is triggered. The multi-level alarm strategy includes sending an alarm SMS to the rental platform and pushing reminders to the user's APP, and freezing the rental rights until manual review.

10. A car rental management system based on visual monitoring, applying the car rental management method based on visual monitoring according to any one of claims 1 to 9, characterized in that: The system includes: The multimodal data acquisition module includes an order information unit, a positioning unit, and a visual monitoring unit. The order information unit is configured to obtain the rental time, location, user credit score, and historical violation records. The positioning unit is configured to obtain the vehicle's real-time location through a high-precision GPS or Beidou positioning chip. The visual monitoring unit includes an in-vehicle visible light camera, an in-vehicle infrared camera, a wide-angle camera on the vehicle body, and a laser radar installed at a detection point, and is configured to obtain driver behavior data and vehicle damage data. The dynamic risk assessment module is configured to: construct a risk scoring model based on the collected multimodal data to generate a real-time risk index; the reference indicators of the risk scoring model include: order risk, vehicle status and driving behavior; The leasing strategy dynamic adjustment module is configured to adjust the leasing terms according to the real-time risk index, and the adjustment content includes the deposit amount and order allocation logic.

Images