A diesel vehicle exhaust blackness grade real-time online monitoring method and system based on video analysis

Abstract

The application discloses a diesel vehicle exhaust blackness grade real-time online monitoring method and system based on video analysis. The method comprises the following steps: collecting visible light and infrared video streams and performing pretreatment; detecting vehicles and performing pixel-level segmentation on exhaust plumes by using a cascade deep network; extracting spatial domain statistics and texture features, frequency domain wavelet transform features, dynamic time sequence features and infrared temperature features from the segmented areas, and performing self-adaptive correction in combination with environmental light parameters; inputting the multi-dimensional feature vectors into a hierarchical integrated classifier, and outputting a Linggeman blackness grade. The system adopts a three-level architecture of an intelligent sensing terminal, an edge computing node and a cloud management and control center. The application realizes accurate exhaust segmentation through an ExhaustFormer network, and realizes high-precision, strong-robustness and low-delay non-contact online monitoring in combination with multi-dimensional robust features and an integrated classification model, thereby providing an efficient technical means for environmental protection supervision.

Description

Technical Field

[0001] This invention relates to the fields of computer vision, artificial intelligence algorithms and environmental monitoring technology, specifically to a method and system for real-time online monitoring of diesel vehicle exhaust opacity levels based on deep learning and multi-dimensional feature analysis. Background Technology

[0002] Black carbon particulate matter in diesel vehicle exhaust is a significant component of urban air pollution, particularly PM2.5. Currently, the regulation of diesel vehicle exhaust blackness both domestically and internationally primarily relies on the following technologies:

[0003] 1. Artificial Ringelmann opacity method: The observer visually assesses the smoke plume and compares it with a standard Ringelmann chart. This method is highly subjective, greatly affected by ambient lighting, background, and the observer's experience, resulting in low efficiency and untraceable data. Essentially, it is a subjective grading assessment of opacity by the human eye.

[0004] 2. Transmission-based smoke meter method: Such as the AVL439 opacity meter, which calculates opacity by measuring the attenuation rate of the light beam passing through the exhaust gas in the exhaust pipe. Although this method is accurate, it is a contact measurement, requiring vehicle modification or fixed-point installation. The equipment is expensive (hundreds of thousands of yuan per unit) and complex to maintain (requiring regular calibration of the optical platform). It is not suitable for large-scale rapid screening of vehicles on the road, and it is even more difficult to build a wide-area online monitoring network.

[0005] 3. Traditional Image Processing Methods: A method for estimating exhaust plume density based on single-frame image grayscale statistics. This method first segments the exhaust region, then calculates the average grayscale value of the region, and finally maps it to Ringelmann gradations using a lookup table. However, this type of method has fundamental flaws: First, single-frame image information is limited and cannot utilize the dynamic characteristics of exhaust gas diffusion; second, simple grayscale features are extremely sensitive to changes in illumination, camera auto-gain, and background interference, resulting in poor robustness; finally, traditional image segmentation algorithms (such as thresholding and edge detection) are ineffective at segmenting semi-transparent and morphologically varied exhaust plumes, leading to a weak foundation for feature extraction.

[0006] In recent years, deep learning techniques, represented by convolutional neural networks (CNNs), have made breakthroughs in object detection and image segmentation. However, directly applying general models (such as Mask R-CNN) to exhaust gas segmentation still faces challenges: exhaust gases are non-rigid, have blurred boundaries, and exhibit varying contrast with backgrounds such as the sky or road surface. Furthermore, how to extract robust feature vectors from the segmented exhaust gas regions that can distinguish different levels of blackness and establish a high-precision mapping relationship with the Ringelmann gradation remains a problem that current research has not yet adequately solved.

[0007] Therefore, developing an online monitoring system that can accurately segment exhaust gases, extract robust multidimensional features, adapt to complex environments, and meet real-time requirements is of urgent need and great significance for improving the technological and precise level of environmental protection supervision. Summary of the Invention

[0008] (a) Purpose of the invention

[0009] The purpose of this invention is to provide a real-time online monitoring method and system for diesel vehicle exhaust opacity levels based on video analysis, to solve the problems of subjective monitoring, low efficiency, high cost, poor environmental adaptability, and inability to be online in real time in existing technologies. This invention aims to achieve non-contact, automated, high-precision, highly robust, and low-latency exhaust opacity level determination.

[0010] (II) Technical Solution

[0011] To achieve the above objectives, the present invention proposes the following technical solution:

[0012] In a first aspect, the present invention provides a method for real-time online monitoring of the opacity level of diesel vehicle exhaust based on video analysis, characterized by comprising the following steps:

[0013] S1: Multi-source video data acquisition and adaptive preprocessing

[0014] Visible light high-definition cameras and infrared thermal imagers are deployed at monitoring points to simultaneously acquire dual video streams containing the target diesel vehicle. and The video stream is preprocessed, including adaptive white balance based on the gray-world assumption, joint denoising based on temporal median filtering and spatial bilateral filtering, and electronic image stabilization based on feature point matching, to obtain a stable video frame sequence. .

[0015] S2: Precise perception of vehicle and exhaust regions based on cascaded deep networks

[0016] S2.1: Vehicle Detection and Localization: A lightweight YOLOv7 model is used as the primary detector for preprocessed visible light frames. Perform real-time detection to obtain the boundary box of the diesel vehicle. And confidence level. The model is optimized for the rear features of diesel vehicles, and the anchor frame size is preferentially matched to the exhaust pipe area of ​​common diesel vehicles.

[0017] S2.2: Pixel-level segmentation of exhaust plumes: based on vehicle bounding box To guide, from and Cropping out the area of ​​interest at the rear of the vehicle and . will double The input is fed into the ExhaustFormer network, which is specifically designed for exhaust gas separation. The core structure of this network is as follows: Figure 2 As shown, it includes:

[0018] Encoder: Employs a hybrid architecture of CNN and Swin Transformer. The shallow CNN quickly extracts low-level features, while the deep Swin Transformer captures the global contextual relationship between exhaust fumes and the background through a window self-attention mechanism.

[0019] Decoder: By progressive upsampling and skip connections, it fuses features from different stages of the encoder to gradually recover spatial details.

[0020] Dynamic sensing module: This module utilizes three consecutive frames feature The temporal dynamics of exhaust gas diffusion are modeled using a lightweight ConvLSTM unit. The state update formula is:

[0021]

[0022]

[0023]

[0024]

[0025]

[0026] in, For the features of the current frame, and Let the hidden state and cell state represent convolution. Represents convolution. It represents the Hadamah accumulation. This represents the sigmoid activation function. and These are trainable parameters. The output of the DPM is concatenated with the decoder features to enhance the model's ability to perceive instantaneous changes in the exhaust gas morphology.

[0027] The network ultimately outputs a result that is similar to... Binary masks with the same resolution The pixel value of 1 indicates that it belongs to the exhaust plume.

[0028] S3: Multi-dimensional, cross-modal blackness feature quantization and fusion

[0029] For segmentation masks For a defined exhaust gas region, extract multi-dimensional feature vectors from the original image. Specifically, it includes:

[0030] S3.1: Spatial Domain Statistics and Texture Features :

[0031] Gray-scale statistical characteristics: Calculate the exhaust gas region in the visible light frame grayscale mean Standard deviation skewness and kurtosis The calculation formula is as follows:

[0032]

[0033]

[0034]

[0035]

[0036] in, For pixel grayscale values, This represents the total number of pixels in the exhaust gas area.

[0037] Improved Local Texture Features: Calculates multi-scale complete local binary pattern histograms. Compared to traditional LBP, CLBP simultaneously encodes the sign, amplitude, and center pixel value of the intensity difference between local pixels, forming a joint histogram. It has stronger robustness to changes in illumination.

[0038] Spatial distribution characteristics: Calculate the centroid coordinates of the exhaust gas region, the aspect ratio of the circumscribed rectangle, and the shape descriptor based on Zernike moments.

[0039] S3.2: Frequency Domain Transformation Characteristics :

[0040] A two-dimensional discrete wavelet transform was performed on the grayscale image of the exhaust gas region. A single-level decomposition was then performed using the db4 wavelet basis function, yielding four sub-bands: the low-frequency approximate sub-band LL, and the high-frequency detail sub-bands LH, HL, and HH. The energy of each sub-band was calculated. Shannon entropy As a characteristic:

[0041]

[0042]

[0043] in, Sub-band coefficient, This represents the normalized coefficient probability distribution. The LL subband energy reflects the overall blackness, while the HH subband energy reflects fine texture.

[0044] S3.3: Dynamic Timing Characteristics :

[0045] Analyze the mask sequence within a short time window (e.g., 5 frames, approximately 0.17 seconds). :

[0046] Diffusion rate: The rate of change of the exhaust gas region area A(t) between adjacent frames. .

[0047] Morphological stability: Calculate the Descein coefficients of masks between consecutive frames. It is used to measure the transient stability of plume morphology.

[0048] Centroid trajectory: Calculates the moving average velocity and direction angle of the centroid coordinate sequence.

[0049] S3.4: Cross-modal infrared signature :

[0050] From synchronized infrared images Extract the average temperature of the corresponding region. Temperature gradient standard deviation Generally speaking, high-temperature areas are usually associated with high concentrations of exhaust gas that has just been emitted.

[0051] S3.5: Environmental Adaptive Feature Correction:

[0052] Ambient light sensor collects light intensity in real time and color temperature Establish a correction model based on multinomial regression to address the grayscale features most directly affected by illumination (such as...). Compensation is performed. The corrected grayscale mean... Calculated by the following formula:

[0053]

[0054] in, These are the model parameters determined through extensive calibration experiments. Similarly, other photosensitive features are corrected to generate more robust feature vectors. .

[0055] Finally, all features are concatenated to form a comprehensive feature vector:

[0056]

[0057] S4: Intelligent classification of blackness levels based on hierarchical ensemble learning

[0058] High-dimensional feature vectors The input is fed into a two-stage hierarchical ensemble classifier, such as... Figure 3As shown.

[0059] Phase 1: Base Classifier Set. Three heterogeneous base classifiers are designed for parallel processing:

[0060] 1. Gradient Boosting Tree: Excels at handling structured numerical features. and Efficiently model numerical features.

[0061] 2. Kernel Support Vector Machine: Finds the optimal classification hyperplane in a high-dimensional feature space, particularly suitable for processing... and The nonlinear characteristics in.

[0062] 3. Shallow one-dimensional convolutional neural network: rearranges feature vectors into pseudo-sequences and uses its local perception ability to capture local dependencies between features.

[0063] Second stage: Meta-classifier. This involves converting the class probability vectors output by the three base classifiers... The features are concatenated to form a new feature vector. Then, a multilayer perceptron (MLP) is used as a meta-classifier to classify... The learning process is conducted to output the final Ringelmann blackness rating. and confidence level .

[0064] During training, a focus loss function is used to address the class imbalance problem (in reality, there are far more low-saturation vehicles than high-saturation vehicles).

[0065]

[0066] in, This represents the probability that the model predicts the true class. This is the adjustment factor (usually set to 2). This represents the category weight.

[0067] S5: Results Encapsulation, Transmission, and Visualization

[0068] The structured monitoring results {timestamp, license plate information (if linked), vehicle location, blackness level L, confidence score, key feature values, exhaust gas area screenshot} are encapsulated into a specified protocol format (such as JSON). This data is then uploaded to a cloud-based environmental monitoring big data platform via 5G or fiber optic network. This platform enables real-time video streaming, automatic alarm for exceeding standards, multi-dimensional analysis of historical data, generation of emission heat maps, and export of statistical reports.

[0069] Secondly, the present invention provides an online monitoring system for implementing the above method, characterized in that it adopts a three-tier architecture of "intelligent sensing terminal - edge computing node - cloud control center", such as... Figure 4 As shown.

[0070] 1. Intelligent Sensing Terminal: Composed of a high-definition visible light intelligent PTZ camera, an uncooled infrared thermal imager, a mini weather station, and a protective housing, it is installed on a gantry or pole. The visible light camera has a resolution ≥4K, a frame rate ≥25fps, and supports wide dynamic range (HDR) and strong light suppression; the infrared thermal imager has a resolution of 640×512; and the mini weather station is used to measure environmental parameters such as light intensity, temperature, humidity, and wind speed.

[0071] 2. Edge computing node: This is the core computing unit, employing an embedded AI industrial control computer. Its built-in modules include:

[0072] Data access and scheduling module: responsible for pulling, decoding and synchronizing multiple video streams.

[0073] Real-time analysis engine: The complete algorithm model for running steps S2-S4 is encapsulated in Docker containers, inference is accelerated by TensorRT, and the pipeline parallelism of vehicle detection, exhaust gas segmentation, feature extraction and classification is realized.

[0074] Local storage and communication module: caches raw data and results, and uploads the results via 5G CPE or industrial switch.

[0075] 3. Cloud-based management center: Built on microservices and containerization technologies, including:

[0076] Data Lake Service: Receives and stores massive amounts of data from monitoring points.

[0077] AI model management service: Supports remote OTA upgrades and A / B testing of edge node models.

[0078] Business application services: Provides web applications such as real-time monitoring, data analysis, device management, and user permissions.

[0079] Visualization service: Display regional emission status through a large-screen cockpit.

[0080] Furthermore, the method also includes a system self-calibration and optimization mechanism:

[0081] Background adaptive modeling: The background model is updated periodically (e.g., hourly) during periods without vehicles to adapt to gradual changes in lighting and weather.

[0082] Online model fine-tuning: For samples with a confidence score in the fuzzy range (e.g., 0.4-0.6), a manual review process is automatically triggered. The reviewed samples are added to the training set, triggering incremental learning of the edge model and enabling continuous evolution of model performance.

[0083] (III) Beneficial Effects

[0084] Compared with the prior art, the solution provided by the present invention has the following outstanding advantages:

[0085] 1. Significantly Improved Monitoring Accuracy: The ExhaustFormer network enables precise segmentation of exhaust gases, laying a reliable foundation for feature extraction. The proposed multi-dimensional feature fusion method (spatial, frequency domain, temporal, and infrared) characterizes blackness from multiple physical perspectives and incorporates environmental adaptive correction, making the features insensitive to changes in illumination and weather. Finally, a hierarchical ensemble classifier makes robust decisions, achieving an overall blackness level classification accuracy of over 96%, far exceeding traditional image processing methods.

[0086] 2. Extremely robust to the environment: Dual-light (visible light + infrared) collaboration, dynamic feature analysis, and a well-defined environmental interference correction model jointly ensure the stable and reliable operation of the system in various complex scenarios such as strong daylight, backlight at dusk, night, light fog, and light rain.

[0087] 3. Excellent real-time performance: Comprehensive optimizations are made from the algorithm level (lightweight model, adaptive frame sampling, pipelined parallelism), hardware level (edge ​​AI computing unit) to the architecture level (end-edge collaboration) to ensure that the end-to-end latency from video input to result output is less than 500 milliseconds, meeting the real-time online monitoring requirements of high-speed vehicles.

[0088] 4. Systematization and Scalability: Provides a complete system solution from smart terminals and edge computing to cloud analytics. The modular design facilitates functional expansion (such as adding a NOx ultraviolet imaging monitoring module), and the cloud-edge collaborative architecture supports unified management and data analysis of tens of thousands of monitoring points, providing a solid technical foundation for building a city-level three-dimensional exhaust emission monitoring network.

[0089] 5. Low deployment and maintenance costs: Non-contact measurement requires no modification to roads or vehicles, and the cost of a single device is much lower than that of a transmission smoke meter station. Furthermore, remote operation and maintenance can significantly reduce the frequency and cost of on-site maintenance. Attached Figure Description

[0090] Figure 1 The overall flowchart of the monitoring method provided in the embodiments of the present invention is shown.

[0091] Figure 2 This is a schematic diagram of the ExhaustFormer exhaust gas separation network structure used in this invention.

[0092] Figure 3 This is a schematic diagram of the hierarchical integrated classifier structure used in this invention.

[0093] Figure 4 This is a schematic diagram of the three-level hardware architecture of the monitoring system provided in an embodiment of the present invention. Detailed Implementation

[0094] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are for illustrative purposes only and do not constitute a limitation on the scope of protection of this invention.

[0095] Example 1: Fixed-point monitoring of urban main roads

[0096] The system of this invention was deployed on a gantry on a main ring road in a certain city.

[0097] Hardware deployment:

[0098] Intelligent sensing terminal: It uses an 8-megapixel high-definition fog-penetrating camera (visible light) and a 640×512 resolution infrared thermal imager, installed side by side, with a field of view covering three lanes. It is equipped with a light sensor.

[0099] Edge computing node: It uses a Jetson AGX Orin 64GB kit, installed in a waterproof enclosure next to the gantry, to power the camera via PoE++ and connect to the Internet via 5GCPE.

[0100] Cloud-based management center: Deployed on the municipal cloud platform, it has an independent server cluster and database.

[0101] Software process:

[0102] 1. The camera captures video at 25fps, and the video stream is directly pushed to the RTSP server on the edge node.

[0103] 2. Edge node running algorithm container:

[0104] a. Preprocessing: Adaptive histogram equalization is performed on the visible light video to enhance contrast.

[0105] b. Detection and Segmentation: Each frame of image is fed into a lightweight YOLOv7 model. After a vehicle is detected, the ExhaustFormer model accelerated by TensorRT is called for exhaust gas segmentation. Utilizing the vehicle's detection bounding box information, segmentation is performed only on the area above the rear of the vehicle to reduce computational load.

[0106] c. Feature Extraction: For the segmented exhaust gas region, spatial features, frequency domain features, and dynamic features are calculated simultaneously according to step S3. At the same time, the current illumination value L=12000 lux is read, and the corrected grayscale mean is calculated using pre-stored parameters. .

[0107] d. Classification: Input the feature vector into the deployed HEC model. The final output level is 1, with a confidence score of 0.88.

[0108] 3. The result is encapsulated as a JSON message and sent to the cloud via a 5G network.

[0109] 4. After receiving the data, the cloud platform updates the monitoring status of the location on the digital map in real time. If the level is ≥2, an audible and visual alarm will be automatically triggered, and information about the suspected violating vehicle will be pushed to the law enforcement terminal App.

[0110] Example 2: Mobile Roadside Enforcement Application

[0111] The core sensing terminal (camera) and edge node (rugged portable industrial control computer) of this invention are integrated into a law enforcement vehicle.

[0112] Work mode:

[0113] The law enforcement vehicle was parked on the side of the road, with the camera pointed at vehicles traveling on the road.

[0114] The system algorithm flow is basically the same as in Example 1, but a vehicle tracking module is added. When a diesel vehicle is detected, the SORT algorithm is used to track the vehicle in subsequent frames, ensuring that the vehicle undergoes only one complete "detection-segmentation-feature extraction-classification" process throughout its passage through the field of view, while continuously updating its dynamic features. Finally, the peak or stable value of the darkness level during its passage is taken as the monitoring result for the vehicle.

[0115] In mobile scenarios, lighting and background changes are more drastic. The system filters out missegmentation of exhaust gas areas caused by vehicle bumps or rapid turns by updating the background model more frequently (every 5 minutes) and utilizing the morphological stability index (Dice coefficient) in dynamic features.

[0116] The enforcement results are transmitted back to the command center via 4G / 5G network, and monitoring reports can be printed on-site via vehicle-mounted printers.

[0117] Effect verification:

[0118] To verify the effectiveness of the system of this invention, a comparative test was conducted for one month at an environmental protection inspection station in a certain city. The system of this invention and a standard transmission smoke meter were used to simultaneously measure 500 diesel vehicles entering the inspection station in the same batch.

[0119] Consistency analysis: The opacity measured by the smoke meter was converted to Ringelmann class according to the standard and compared with the results of this system. The consistency reached 95.6%.

[0120] Efficiency comparison: The average monitoring time per vehicle by the system of this invention is about 2.1 seconds (from entering the field of view to outputting the result), while the smoke meter takes more than 45 seconds to complete one standard measurement cycle, which is more than 20 times more efficient.

[0121] Environmental adaptability: The test included sunny, cloudy, and rainy weather. In sunny, strong backlight conditions, traditional image processing methods failed, while the accuracy of the system in this invention decreased by only about 2% through feature correction.

[0122] The above embodiments fully demonstrate the significant advantages of the method and system of the present invention in terms of accuracy, efficiency, robustness and practicality.

[0123] Industrial applicability of the invention

[0124] The video analysis-based real-time online monitoring method and system for diesel vehicle exhaust opacity levels provided by this invention has high technological maturity and clear prospects for industrial production and large-scale application.

[0125] 1. Mature industry chain: The high-definition cameras, AI computing chips, communication modules, etc. that it relies on are all mature products in the market, which are easy to procure and integrate.

[0126] 2. Standardized production: The system hardware can adopt standard chassis and modular design, and the software can be packaged into standardized Docker images or SDKs, which facilitates mass production and deployment.

[0127] 3. Clear market demand: It meets the urgent need for refined regulation of motor vehicle pollution worldwide, with a clear customer base (environmental protection and transportation management departments in various regions, large transportation companies), and a huge market capacity.

[0128] 4. Compliance: The system design complies with relevant electrical safety, data security and environmental monitoring standards. The measurement methods can be equivalently correlated with the existing Ringelmann visual inspection method and smoke meter method, providing a basis for law enforcement that complies with regulatory requirements.

[0129] Therefore, this invention has good industrial applicability and can generate positive socio-economic benefits.

Claims

1. A method for real-time online monitoring of diesel vehicle exhaust opacity levels based on video analysis, characterized in that, Includes the following steps: S1: Acquire a video stream containing diesel vehicles and preprocess the video stream; S2: Detect diesel vehicle targets from preprocessed video frames and perform pixel-level segmentation of exhaust plumes in the exhaust pipe area of ​​the diesel vehicle to obtain an exhaust region mask. S3: Based on the exhaust gas area mask, extract multi-dimensional blackness feature vectors from the corresponding original image area; The multi-dimensional blackness feature vector includes: spatial domain statistics and texture features, frequency domain transformation features obtained by wavelet transforming the exhaust gas region image, and dynamic temporal features obtained by analyzing exhaust gas region masks in multiple consecutive frames. S4: Input the multi-dimensional blackness feature vector into the pre-trained blackness level classification model and output the corresponding Ringelmann blackness level; S5: Output and upload monitoring results.

2. The method according to claim 1, characterized in that, In step S2, the pixel-level segmentation is implemented using an encoder-decoder network; the encoder includes a convolutional neural network layer and a Swing Transformer module; the decoder includes a dynamic perception module, which performs temporal modeling of intermediate features of multiple consecutive frames through convolutional long short-term memory network units to enhance the perception capability of instantaneous morphological changes of exhaust gas.

3. The method according to claim 2, characterized in that, The state update process of the convolutional long short-term memory network unit in the dynamic perception module includes: in, For the features of the current frame, and for Constantly hide state and cell state, Represents convolution. It represents the Hadamah accumulation. This represents the sigmoid activation function. and These are trainable parameters.

4. The method according to claim 1, characterized in that, In step S3, the spatial domain statistics and texture features include the grayscale mean of the exhaust gas region. Standard deviation skewness kurtosis ; The calculation formulas are as follows: in, For pixel grayscale values, This represents the total number of pixels in the exhaust gas area.

5. The method according to claim 4, characterized in that, Step S3 also includes an environment adaptive feature correction step: acquiring ambient light intensity. Establish a correction model for grayscale mean. Compensation is performed, and the corrected grayscale mean value is obtained. The calculation formula is: in, These are the model parameters determined through calibration.

6. The method according to claim 1, characterized in that, In step S4, the blackness level classification model is a hierarchical ensemble classifier, which includes: a base classifier group, consisting of at least two heterogeneous base classifiers, wherein the base classifiers are selected from at least two of gradient boosting decision trees, support vector machines, and one-dimensional convolutional neural networks; and a meta-classifier, which receives a stacked feature vector composed of the probability vectors output by each of the base classifiers, performs the final classification decision, and outputs the Ringelmann blackness level.

7. The method according to claim 1, characterized in that, It also includes a real-time optimization step: dynamically adjusting the sampling frame rate of video analysis based on vehicle speed, traffic density, and system load at the monitoring points.

8. A real-time online monitoring system for the opacity level of diesel vehicle exhaust gases based on video analysis, used to implement the method according to any one of claims 1 to 7, characterized in that, The system adopts a three-tier architecture consisting of intelligent sensing terminals, edge computing nodes, and a cloud-based control center. The intelligent sensing terminals include visible light high-definition cameras. The edge computing nodes have built-in AI computing units for running vehicle detection, exhaust gas segmentation, feature extraction, and classification algorithm models. The cloud-based control center is used to receive, store, analyze, and display the monitoring results.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images