A method, device, and storage medium for monitoring vehicle driving behavior.

Abstract

This invention discloses a method and device for monitoring vehicle driving behavior. The monitoring method includes collecting vehicle driving information and surrounding environmental information; preprocessing the collected driving information and surrounding environmental information; fusing the preprocessed driving information and surrounding environmental information to form training samples; constructing a neural network model of the vehicle based on the vehicle's basic information and physical engine; training the neural network model using the training samples; predicting vehicle driving behavior using the trained neural network model; and adaptively adjusting the frame rate between the predicted results and the actual results using the vehicle's physical engine, and displaying the adjustment results to the user. This method can predict the consequences of various driving behaviors, and has high data playback accuracy and excellent visual effects.

Description

Technical Field

[0001] This invention relates to the field of driving monitoring, and in particular to a method and apparatus for monitoring driving behavior on racetracks or test tracks. Background Technology

[0002] In racing tracks or vehicle performance testing grounds, it is necessary to replay vehicle driving data. Existing vehicle driving data replay solutions can only replay actual data and cannot predict the consequences of other driving behaviors, which is not conducive to analysis and summarization. Moreover, existing replays all use a fixed frame rate timeline linear playback mechanism. This type of solution has obvious technical shortcomings: First, relying solely on timestamps to control the playback rhythm makes it difficult to restore the vehicle's dynamic characteristics, such as the physical features of rapid acceleration and steering, resulting in discrepancies between the replay and the actual driving process; Second, it lacks adaptive adjustment capabilities based on dynamic constraints, and cannot use physical laws to fill in frames when faced with missing data or complex conditions, affecting the accuracy of driving behavior analysis; Third, the allocation of computing resources is not reasonable, and it is prone to stuttering or resource waste when running on different performance devices. Summary of the Invention

[0003] To address the aforementioned problems, this invention provides a monitoring solution that can predict the consequences of various driving behaviors, and offers high data playback accuracy and excellent visual effects.

[0004] To achieve the above objectives, the present invention provides a method for monitoring vehicle driving behavior, comprising: collecting vehicle driving information, including basic vehicle information and driving information, wherein the driving information includes one or more of the following: speed, steering wheel angle, accelerator pedal position, brake pedal travel, and gear position signal; collecting information about the vehicle's surrounding environment, including location information, lane line information, and traffic light information; preprocessing the collected driving information and surrounding environment information; fusing the preprocessed driving information and surrounding environment information to form training samples; constructing a neural network model of the vehicle based on the vehicle's basic information and the vehicle's physics engine; training the neural network model using the training samples; predicting vehicle driving behavior using the trained neural network model; and adaptively adjusting the frame rate between the predicted results and the actual results using the vehicle's physics engine, and displaying the adjustment results to the user.

[0005] Optionally, the collected driving information and vehicle surrounding environment information are preprocessed, including: synchronizing the collected driving information and vehicle surrounding environment information in time; establishing a three-dimensional coordinate system with the center of the vehicle's rear axle as the origin, unifying all information involving coordinate positions into spatial coordinates; and using a Kalman filter to filter out abnormal data.

[0006] Optionally, the preprocessed driving information and vehicle surrounding environment information are fused to form training samples, including: extracting feature vectors from the preprocessed driving information and vehicle surrounding environment information; normalizing the feature vectors; identifying outliers using the 3σ principle and replacing them with linear interpolations of nearby normal values; and using principal component analysis to compress the dimensions of similar feature vectors.

[0007] Optionally, a neural network model of the vehicle is constructed based on the vehicle's basic information and the vehicle's physics engine, including: configuring the vehicle model based on the vehicle's basic information; including vehicle rigid body properties, tire model, and suspension system model; calculating the vehicle's motion state based on the vehicle's driving information; constructing a causal inference model and optimizing the causal inference model using the vehicle model and the vehicle's motion state.

[0008] Optionally, a causal inference model is constructed, including: constructing a causal graph of driving behavior; simulating intervention operations using Do-calculus to generate counterfactual trajectories; and validating the prediction results using a counterfactual error model, wherein the counterfactual error model is as follows:

[0009] ;

[0010] Where CFE represents counterfactual error, N represents the number of samples, and i represents the i-th counterfactual scenario. This indicates the prediction result for a counterfactual scenario. It represents the true outcome of a counterfactual scenario.

[0011] Optionally, the frame rate can be adaptively adjusted between the predicted and actual results using the vehicle's physics engine, and the adjustment results can be displayed to the user. This includes: obtaining position and speed information from the predicted and actual results; calculating the position difference and speed difference between the predicted and actual results respectively; calculating the frame rate adjustment value using a PID algorithm; and adjusting the frame rate of the displayed results according to the frame rate adjustment value.

[0012] Optionally, adjusting the frame rate of the displayed result based on the frame rate adjustment value includes: controlling the playback frame rate of the actual result through a master timer; controlling the playback frame rate of the predicted result through a physical timer; calculating the time that the master timer needs to be adjusted based on the frame rate value; and adjusting the time of the master timer based on the calculation result.

[0013] Optionally, the feature is that the time that the master timer needs to be adjusted is calculated using the following formula.

[0014]

[0015] Where T is the time interval of the main timer, and f is the frame rate adjustment value, which is dynamically generated by the PID algorithm.

[0016] On the other hand, the present invention also provides a vehicle driving behavior monitoring device, comprising: a first acquisition unit for acquiring vehicle driving information, including basic vehicle information and driving information, wherein the driving information includes one or more of speed, steering wheel angle, accelerator pedal position, brake pedal travel, and gear signal; a second acquisition unit for acquiring vehicle surrounding environment information, including location information, lane line information, and traffic light information; a preprocessing unit for preprocessing the acquired driving information and vehicle surrounding environment information; a feature fusion unit for fusing the preprocessed driving information and vehicle surrounding environment information to form training samples; a construction unit for constructing a neural network model of the vehicle based on the vehicle's basic information and the vehicle's physics engine; a training unit for training the neural network model using the training samples; a prediction unit for predicting vehicle driving behavior using the trained neural network model; and an output unit for adaptively adjusting the frame rate of the prediction result and the actual result through the vehicle's physics engine, and displaying the adjustment result to the user.

[0017] The advantages of this invention over the prior art are:

[0018] This solution uses a vehicle physics engine to build a neural network model to predict the consequences of various driving behaviors in a given scenario. It also compares the predicted results with the actual results, providing the most direct and effective reflection of vehicle test or competition data, which is beneficial for data analysis and summarization. In addition, a vehicle dynamics physics engine is introduced into the replay timer mechanism to dynamically adjust the timer interval based on parameters such as acceleration and steering angle of the current replay frame (e.g., shortening the 100ms interval to 80ms in a rapid acceleration scenario), ensuring that the vehicle speed changes during replay are consistent with the dynamic characteristics of actual driving. Attached Figure Description

[0019] Figure 1 This is a flowchart of a vehicle driving behavior monitoring method provided by the present invention;

[0020] Figure 2 This is a structural diagram of a vehicle driver behavior monitoring device provided by the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0022] Reference Figure 1This embodiment provides a method for monitoring vehicle driving behavior, including the following steps:

[0023] S10: Collect vehicle driving information.

[0024] Specifically, in this embodiment, parameters can be collected via the CAN bus, including but not limited to: speed, longitudinal acceleration, steering wheel angle, accelerator pedal position, brake pedal travel, gear signal, etc.

[0025] In addition, data such as engine speed, intake manifold pressure, coolant temperature, and oxygen sensor voltage can be collected through the OBD interface.

[0026] S20: Collect information about the environment surrounding the vehicle.

[0027] Specifically, in this embodiment, information about the vehicle's surroundings, including lane lines and even traffic lights, is collected via an onboard camera, and the vehicle's current location information is collected via GPS.

[0028] S30: Preprocessing of collected driving information and vehicle surrounding environment information.

[0029] Specifically, the collected driving information and vehicle surrounding environment information are first synchronized in time. In this embodiment, NTPv4 (Network Time Protocol) is used to obtain UTC standard time through a 4G module to achieve nanosecond-level time alignment of multiple source sensors. Specific measures include: CAN bus synchronization: sending a synchronization frame with ID=0x100 every 100ms, containing a timestamp accurate to microseconds; camera synchronization: embedding a PTP (Precise Time Protocol) timestamp into each frame of video stream; radar hardware synchronization: achieving synchronization through GPIO hardware trigger signals; clock drift compensation: using a Kalman filter algorithm to dynamically correct the clock deviation of each device.

[0030] Next, a three-dimensional coordinate system with the center of the vehicle's rear axle as the origin is established, and all information involving coordinate position is unified into spatial coordinates. In this embodiment, a three-dimensional coordinate system with the center of the vehicle's rear axle as the origin is established (X-axis forward, Y-axis left, Z-axis upward), and then the data collected by each sensor is transformed into spatial coordinates to establish a unified spatial coordinate system.

[0031] Finally, a Kalman filter is used to filter out abnormal data. In this embodiment, a Kalman filter model is used to reduce noise in the GPS trajectory; and dynamic consistency verification is performed, including establishing velocity-displacement constraints, acceleration continuity constraints, and steering rationality constraints. For points that do not meet the above three constraints, cubic spline interpolation is used to correct the abnormal data.

[0032] S40: The preprocessed driving information and the information about the vehicle's surrounding environment are fused to form training samples.

[0033] Specifically, in this embodiment, linear interpolation is used to align the time of data from different sampling rates. Then, feature vectors of the preprocessed driving information and the vehicle's surrounding environment information are extracted. In this embodiment, 128-dimensional feature vectors are extracted from the preprocessed data, including 9 categories: vehicle motion features (6 dimensions): real-time speed, acceleration, jerk value, steering angular velocity, gear, wheel speed difference; control input features (6 dimensions): steering wheel angle, steering rate, throttle opening, brake pressure, clutch travel, handbrake status; environmental interaction features (8 dimensions): distance to the vehicle in front, distance to the vehicle to the side, lane departure, lane curvature, traffic light status, pedestrian distance, number of obstacles, road slope; driver state features (8 dimensions): gaze deviation angle, blink frequency, head pitch angle, grip strength, release time, heart rate variability, facial expression entropy, voice emotion index; dynamic features (8 dimensions): longitudinal G value, lateral G value, etc. Features include: pitch angle, roll angle, tire slip ratio, brake force distribution, driving torque, and suspension travel; time-domain derived features (16 dimensions): acceleration / deceleration volatility, steering smoothness, throttle change rate, braking frequency, following distance, lane keeping time, lane change frequency, and overtaking frequency; frequency-domain derived features (16 dimensions): steering wheel angle spectral energy, acceleration spectral entropy, throttle signal main frequency, and brake signal bandwidth; scene features (16 dimensions): weather type, light intensity, road type, traffic density, time of day, and geographical location; and prediction and comparison features (30 dimensions): short-term trajectory prediction, conflict probability prediction, historical behavior bias, and expert operation similarity.

[0034] Next, the feature vector is normalized, and outliers are identified using the 3σ principle. The outliers are then replaced with linear interpolations of the nearest normal values.

[0035] Finally, principal component analysis is used to compress the dimensions of similar feature vectors. In this embodiment, vectors with a correlation coefficient > 0.9 are subjected to dimension compression.

[0036] S50: Construct a neural network model of the vehicle based on the vehicle's basic information and physical engine.

[0037] By constructing a digital twin of a vehicle, high-precision simulation of the vehicle's physical behavior can be achieved, providing a foundation for subsequent behavior analysis and prediction.

[0038] Specifically, this embodiment uses the NVIDIA PhysX 5.0 physics engine as the core computing platform for digital twins, which has efficient rigid body dynamics and collision detection capabilities, making it suitable for real-time vehicle simulation.

[0039] First, the vehicle model is configured based on the vehicle's basic information, including the vehicle's rigid body properties, tire model, and suspension system model. The tire model uses the Pacejka magic formula to establish the tire mechanics model.

[0040] Next, the vehicle's motion state is calculated based on the vehicle's driving information, including longitudinal dynamics calculation, lateral dynamics calculation, aerodynamic effects, and the application of forces and torques. That is, based on the calculated forces and torques, they are applied to the vehicle's rigid body through the physics engine API to update its position and attitude.

[0041] Finally, a causal inference model is constructed and optimized using the vehicle model and vehicle motion state.

[0042] In this embodiment, the causal inference model is constructed by first building a causal graph of driving behavior. It should be noted that causal relationships include direct causal relationships and indirect causal relationships. Direct causal relationships, such as steering wheel angle, directly affect the vehicle trajectory.

[0043] Next, in the physics engine, intervention operations are simulated using Do-calculus (such as forcibly setting the steering wheel angle to \(\theta + 10^\circ\)) to generate counterfactual trajectories.

[0044] Finally, the prediction results are verified using a counterfactual error model, which is as follows:

[0045] ;

[0046] Where CFE represents counterfactual error, N represents the number of samples, and i represents the i-th counterfactual scenario. This indicates the prediction result for a counterfactual scenario. It represents the true outcome of a counterfactual scenario.

[0047] Experiments show that after introducing causal inference, the counterfactual prediction error decreased from 1.2 to 0.7, and the prediction lead time increased from 3 seconds to 5 seconds.

[0048] S60: Train the neural network model using the training samples.

[0049] S70: Predict vehicle driving behavior using a trained neural network model.

[0050] S80: Adaptively adjusts the frame rate between the predicted and actual results using the vehicle's physics engine, and displays the adjusted results to the user.

[0051] Specifically, it acquires the position and velocity information from the predicted results and the actual results; it obtains the current simulated position (pos) and velocity (vel) in real time from the vehicle dynamics model, which is generated by the CAN data-driven physics engine (such as the result of the accelerator, brake and other inputs calculated by the dynamic equation).

[0052] Next, the position and velocity differences in the predicted and actual results are calculated separately. The position error represents the difference between the position simulated by the physics engine and the target position, reflecting the degree of positional deviation in the current playback frame. The velocity error represents the difference between the velocity simulated by the physics engine and the target velocity, reflecting the degree of deviation in motion speed.

[0053] Then, the PID algorithm is used to calculate the frame rate adjustment value.

[0054] Next, the frame rate displayed based on the frame rate adjustment value is adjusted, including: controlling the playback frame rate of the actual result through the master timer; controlling the playback frame rate of the predicted result through the physical timer; and calculating the adjustment time required for the master timer using the following formula.

[0055]

[0056] Where T is the time interval of the main timer, and f is the frame rate adjustment value, which is dynamically generated by the PID algorithm.

[0057] Finally, adjust the master timer's time based on the calculation results.

[0058] This adaptive frame rate method ensures that during playback, in rapidly accelerating scenes where the physics engine's calculated speed is lower than the target speed, `pos_error` becomes positive, and the proportional term increases the frame rate, accelerating the playback and quickly catching up to the target state. Simultaneously, the differential term detects a rapid increase in error and proactively increases its adjustment to avoid latency. In constant-speed driving scenes, the error approaches zero, and the three adjustment terms balance each other, stabilizing the frame rate at a base value (e.g., 30fps) to guarantee smooth gameplay.

[0059] This solution couples the dynamic characteristics simulated by the physics engine with the playback frame rate through a PID algorithm, making the motion states of the screen (such as acceleration jerks and the feeling of centrifugal force during steering) consistent with real driving. Moreover, it does not require preset frame rate rules, but dynamically adjusts in real time according to driving behavior, and is compatible with complex conditions such as rapid acceleration and emergency braking.

[0060] Reference Figure 2 This embodiment also provides a vehicle driving behavior monitoring device, including:

[0061] The first acquisition unit 100 is used to acquire vehicle driving information, including basic vehicle information and driving information, wherein the driving information includes one or more of the following: speed, steering wheel angle, accelerator pedal position, brake pedal travel, and gear signal. It should be noted that since the specific acquisition method and process have been described in detail in step S10 of the above-mentioned vehicle driving behavior monitoring method, they will not be repeated here.

[0062] The second acquisition unit 200 is used to acquire information about the vehicle's surrounding environment, including location information, lane line information, and traffic light information. It should be noted that since the specific acquisition method and process have been described in detail in step S20 of the above-mentioned method for monitoring vehicle driving behavior, they will not be repeated here.

[0063] The preprocessing unit 300 is used to preprocess the collected driving information and vehicle surrounding environment information. It should be noted that since the specific preprocessing method and process have been described in detail in step S30 of the above-mentioned vehicle driving behavior monitoring method, they will not be repeated here.

[0064] The feature fusion unit 400 is used to fuse the preprocessed driving information and the vehicle's surrounding environment information to form training samples. It should be noted that since the specific feature fusion method and process have been described in detail in step S40 of the above-mentioned vehicle driving behavior monitoring method, they will not be repeated here.

[0065] The construction unit 500 is used to construct a neural network model of the vehicle based on the vehicle's basic information and the vehicle's physical engine. It should be noted that since the specific construction method and process have been described in detail in step S50 of the above-mentioned method for monitoring vehicle driving behavior, they will not be repeated here.

[0066] The training unit 600 is used to train the neural network model using the training samples. It should be noted that since the specific training method and process have been described in detail in step S60 of the above-mentioned vehicle driving behavior monitoring method, they will not be repeated here.

[0067] The prediction unit 700 is used to predict vehicle driving behavior using a trained neural network model. It should be noted that since the specific prediction method and process have been described in detail in step S70 of the above-mentioned vehicle driving behavior monitoring method, they will not be repeated here.

[0068] The output unit 800 is used to adaptively adjust the frame rate of the predicted and actual results using the vehicle's physics engine, and then display the adjustment results to the user. It should be noted that the specific output method and process have been described in detail in step S10 of the aforementioned vehicle driving behavior monitoring method, and therefore will not be repeated here.

[0069] In addition, embodiments of the present invention also provide a computer-readable storage medium, wherein the computer-readable storage medium may store a program that, when executed, includes some or all of the steps of the method for monitoring any vehicle driving behavior described in the above method embodiments.

[0070] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0071] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0072] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage device, which may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0073] The above description, with reference to the accompanying drawings, illustrates an exemplary flowchart for monitoring the driving behavior of a vehicle according to an embodiment of the present invention. It should be noted that the numerous details included in the above description are merely illustrative of the invention and not intended to limit it. In other embodiments of the invention, the method may have more, fewer, or different steps, and the order, inclusion, function, and other relationships between the steps may differ from those described and illustrated.

Claims

1. A method for monitoring vehicle driving behavior, characterized in that, include: Collect vehicle driving information, including basic vehicle information and driving information, including speed, steering wheel angle, accelerator pedal position, brake pedal travel, and gear signal; Collect information about the vehicle's surrounding environment, including location information, lane markings, and traffic light information; Preprocessing of collected driving information and vehicle surrounding environment information; The preprocessed driving information and the vehicle's surrounding environment information are fused to form training samples; A neural network model of the vehicle is constructed based on the vehicle's basic information and physical engine, including configuring the vehicle model based on the vehicle's basic information; including vehicle rigid body properties, tire model, and suspension system model; calculating the vehicle's motion state based on the vehicle's driving information; constructing a causal inference model and optimizing the causal inference model using the vehicle model and the vehicle's motion state; The neural network model is trained using the training samples. Predict vehicle driving behavior using a trained neural network model; The vehicle's physics engine adaptively adjusts the frame rate of the predicted and actual results and displays the adjustment results to the user, including obtaining position and speed information from the predicted and actual results. Calculate the position difference and velocity difference in the predicted results and the actual results respectively; The frame rate adjustment value is calculated using the PID algorithm. The frame rate displayed is adjusted based on the frame rate adjustment value. The process involves constructing a causal inference model, including building a causal graph of driving behavior; simulating intervention operations using Do-calculus to generate counterfactual trajectories; and validating the prediction results using a counterfactual error model, which is as follows: ； Where CFE represents counterfactual error, N represents the number of samples, and i represents the i-th counterfactual scenario. This indicates the prediction result for a counterfactual scenario. This represents the true outcome of a counterfactual scenario; The adjustment of the displayed frame rate based on the frame rate adjustment value includes controlling the playback frame rate of the actual result through the master timer; controlling the playback frame rate of the prediction result through the physical timer; calculating the time that the master timer needs to adjust based on the frame rate adjustment value; adjusting the time of the master timer based on the calculation result; the time that the master timer needs to adjust is calculated using the following formula; ； Where T is the time interval of the main timer, and f is the frame rate adjustment value, which is dynamically generated by the PID algorithm.

2. The monitoring method according to claim 1, characterized in that, Preprocessing of collected driving information and vehicle surrounding environment information includes: The collected driving information and information about the vehicle's surrounding environment are synchronized in time. Establish a three-dimensional coordinate system with the center of the vehicle's rear axle as the origin, and unify all information involving coordinate positions into spatial coordinates; Use a Kalman filter to filter out abnormal data.

3. The monitoring method according to claim 1, characterized in that, The preprocessed driving information and the vehicle's surrounding environment information are fused to form training samples, including: Extract feature vectors from the preprocessed driving information and the vehicle's surrounding environment information; The feature vectors are normalized. Outliers are identified using the 3σ principle and replaced with linear interpolations of nearby normal values. Principal component analysis is used to compress the dimensionality of similar feature vectors.

4. A device for monitoring vehicle driving behavior, characterized in that, include: The first acquisition unit is used to collect the vehicle's driving information, including the vehicle's basic information and driving information, wherein the driving information includes one or more of the following: speed, steering wheel angle, accelerator pedal position, brake pedal travel, and gear signal. The second data acquisition unit is used to collect information about the vehicle's surrounding environment, including location information, lane line information, and traffic light information. The preprocessing unit is used to preprocess the collected driving information and vehicle surrounding environment information. The feature fusion unit is used to fuse preprocessed driving information and vehicle surrounding environment information to form training samples. The construction unit is used to construct a neural network model of the vehicle based on the vehicle's basic information and physical engine. This includes configuring the vehicle model based on the basic information, including vehicle rigid body properties, tire models, and suspension system models; calculating the vehicle's motion state based on its driving information; constructing a causal inference model and optimizing the causal inference model using the vehicle model and the vehicle's motion state. The construction of the causal inference model includes constructing a causal graph of driving behavior; simulating intervention operations using Do-calculus to generate counterfactual trajectories; and verifying the prediction results using a counterfactual error model, wherein the counterfactual error model is as follows: ； Where CFE represents counterfactual error, N represents the number of samples, and i represents the i-th counterfactual scenario. This indicates the prediction result for a counterfactual scenario. This represents the true outcome of a counterfactual scenario; Training unit, used to train the neural network model using the training samples; The prediction unit is used to predict vehicle driving behavior using a trained neural network model. The output unit is used to adaptively adjust the frame rate of the predicted and actual results using the vehicle's physics engine, and display the adjustment results to the user. This includes acquiring position and speed information from the predicted and actual results; calculating the position and speed differences between the predicted and actual results; calculating the frame rate adjustment value using a PID algorithm; and adjusting the displayed frame rate based on the frame rate adjustment value. The adjustment of the displayed frame rate based on the frame rate adjustment value includes controlling the playback frame rate of the actual results using a master timer; controlling the playback frame rate of the predicted results using a physical timer; calculating the adjustment time required by the master timer based on the frame rate adjustment value; and adjusting the master timer's time based on the calculation result. The adjustment time required by the master timer is calculated using the following formula. ； Where T is the time interval of the main timer, and f is the frame rate adjustment value, which is dynamically generated by the PID algorithm.

5. A computer-readable storage medium, comprising a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of a vehicle driving behavior monitoring method as described in any one of claims 1 to 3.

Images