Garbage compression intelligent control system

Abstract

The invention discloses an intelligent control system for garbage compression, and relates to the technical field of control, a box body detection module selects to notify transfer or to wait or to compare on the basis of detecting the triggering opportunity and whether a box body exists or not; after the identity verification is successful, the identification judgment module detects whether a box body exists or not, if the identification judgment module is notified to wait, if the identification judgment module is notified to compare, whether a waiting garbage truck exists or not is judged, if not, the identification judgment module determines the residual volume of the box body to judge whether the box is full or not, if yes, the identification judgment module informs transfer, and if not, the identification judgment module measures the volume of carried garbage to decide full unloading or separate unloading; the compression execution module judges whether the box is full or not again after executing full unloading, if yes, transshipment is notified, if not, waiting is carried out, and if a separate unloading instruction is received, compression is carried out after the box is full, and comparison is notified again; and the transfer processing module sends out a transfer notice and notifies comparison after the empty box is in place, so that efficient processing of the garbage compression process is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of control, in particular to a garbage compression intelligent control system. BACKGROUND

[0002] As a key link in the garbage disposal process, the intelligent degree of garbage compression directly affects the garbage transfer efficiency, transportation cost and terminal processing pressure. Building a garbage compression control system with automatic detection, accurate decision and efficient execution capability has become an important breakthrough to improve the overall efficiency of urban environmental sanitation system and reduce operating costs.

[0003] The existing Chinese patent application with publication number CN111352368A discloses a control method, device and system for a garbage compression station. The network control system receives control instructions from the target object. The garbage compression control system controls the working mode of the garbage compression station according to the control instructions, and in the case that the working mode is the garbage compression mode, the package length information and the single garbage weight of the garbage package falling into the garbage compressor are obtained, as well as the total weight of the garbage in the garbage compressor. According to the package length information, the single garbage weight and the total weight of the garbage, the garbage compressor is controlled to compress the garbage package, solving the technical problem of low garbage compression processing efficiency.

[0004] The existing garbage compression control system has the following problems: first, the garbage dumping and transfer link lacks intelligent identity verification and process coordination mechanism, and there are management loopholes such as illegal dumping and vehicle scheduling confusion; second, the data interaction and decision logic of different links are disconnected, and the whole chain closed-loop control from detection, compression to transfer cannot be realized, resulting in low efficiency and resource waste of the whole processing flow. SUMMARY

[0005] The present application proposes a garbage compression intelligent control system to provide integrated closed-loop control of box changing and unloading with multi-modal identity verification and multi-module data interaction, and to realize efficient garbage compression.

[0006] The technical solution to achieve the purpose of the present application is as follows:

[0007] A garbage compression intelligent control system includes a box detection module, an identification and decision module, a compression execution module and a transfer processing module.

[0008] The box detection module scans and obtains the point cloud in the cavity when the system starts or receives the box detection instruction and performs preprocessing, determines the regular contour through edge detection, contour tracking and convex hull calculation, calculates the geometric shape features and performs rule verification, and selects to send the transfer instruction or feedback waiting instruction or feedback comparison instruction;

[0009] The identification and judgment module uses visual recognition technology to identify the garbage truck model and license plate, combines NFC recognition for identity verification, and sends a bin inspection command upon successful verification. If a waiting command is received, it waits; if a comparison command is received, it selects to wait or deletes the waiting command based on its existence, and determines the remaining volume of the bin using infrared ranging. It then decides whether to send transfer instructions or use a millimeter-wave radar array for scanning, and constructs a projection data distribution based on radar ranging principles and echo signals. The attenuation coefficient distribution is then updated using a hierarchical iterative algorithm. The volume of the waste carried was obtained based on grayscale conversion and three-dimensional topological segmentation. Compare the remaining volume of the box With the volume of garbage carried The decision is made to send either a full unloading command or a partial unloading command, wherein, and These are the installation angle and radial distance, respectively. Represents spatial coordinates;

[0010] The compression execution module receives the full unloading command and continuously monitors the remaining volume of the enclosure using infrared ranging. Changes will continue until the remaining volume of the enclosure remains within the preset time period. Compress the container when there is no change, and then determine the remaining volume of the container again after compression. and with volume threshold The system compares data and decides whether to send a transfer instruction or wait based on the comparison results. It also receives unloading instructions and continuously monitors the remaining volume of the container using infrared ranging. until the remaining volume of the box equal to volume threshold When the time comes, a stop-pour prompt is sent and compression is initiated; after compression, a comparison command is sent back.

[0011] When the transfer processing module receives a transfer instruction, it issues a transfer notification, raises the carrying platform, and determines whether the empty container is in place by coordinating the status of the limit switch and the weight sensor. After the empty container is in place, it identifies the transfer vehicle model and license plate through visual recognition technology, and performs identity verification by combining NFC recognition. After successful verification, it lowers the carrying platform and sends back a comparison instruction. The visual recognition technology, identity verification method and identification decision module are consistent.

[0012] Furthermore, the enclosure detection module includes a detection triggering unit, a detection analysis unit, and a processing feedback unit;

[0013] When the system starts up or receives a test command, the detection trigger unit performs a laser scan on the cavity to acquire the point cloud inside the cavity. ;

[0014] The detection and analysis unit analyzes the intracavity point cloud. Down-sampling and filtering are performed, edge points are identified based on edge detection, and a regular contour is determined using contour tracking and convex hull calculation, rectangularity and height are calculated, and rule verification is performed;

[0015] The processing feedback unit sends a transfer instruction when the detection trigger is system startup and the rule verification result is no box, waits when the detection trigger is system startup and the rule verification result is a box, feeds back a waiting instruction when the detection trigger is receiving a box detection instruction and the rule verification result is no box, and feeds back a comparison instruction when the detection trigger is receiving a box detection instruction and the rule verification result is a box.

[0016] Specifically, the detection and analysis unit obtains the intracavity point cloud Then, the three-dimensional space is divided into cubic voxels, the centroid point coordinates of the intracavity point coordinates in each cubic voxel are calculated and replace the intracavity point coordinates in the cubic voxel, and the centroid point cloud is generated by down-sampling .

[0017] Further, the centroid point cloud is processed using bilateral filtering , the neighborhood of the first centroid point coordinate in the centroid point cloud is searched and the covariance matrix of the neighborhood is calculated , the covariance matrix of the neighborhood is feature decomposed, the eigenvector corresponding to the smallest eigenvalue is selected as the normal vector of the first centroid point coordinate , the spatial distance between each centroid point coordinate in the neighborhood and the first centroid point coordinate is calculated and the similarity weight is calculated , the distance weight and the similarity weight of each centroid point coordinate in the neighborhood are weighted and summed and divided by the corresponding weight product, to obtain the first denoising point coordinate , each centroid point coordinate in the centroid point cloud is bilateral filtered to generate the denoising point cloud , . . , .

[0018] ​​​​​​Further, the detection analysis unit calculates the gradient amplitude and gradient direction of the first denoised point coordinate by using the Sobel operator The gradient amplitude and gradient direction of the first denoised point coordinate The gradient amplitude and gradient direction of the first denoised point coordinate The gradient amplitude and gradient direction of the first denoised point coordinate The gradient amplitude and gradient direction of the first denoised point coordinate The gradient amplitude and gradient direction of the first denoised point coordinate The gradient amplitude and gradient direction of the first denoised point coordinate The gradient amplitude and gradient direction of the first denoised point coordinate The gradient amplitude and gradient direction of the first denoised point coordinate The gradient amplitude and gradient direction of the first denoised point coordinate The gradient amplitude and gradient direction of the first denoised point coordinate The gradient amplitude and gradient direction of the first denoised point coordinate The gradient amplitude and gradient direction of the first denoised point coordinate The gradient amplitude and gradient direction of the first denoised point coordinate .

[0019] Further, based on contour tracking, an untracked strong edge point in the edge point cloud is selected, and adjacent edge points are sequentially searched in a clockwise direction based on neighborhood search until the starting strong edge point is returned or a new connected edge point cannot be searched, thereby obtaining a continuous contour line. The process is repeated until all strong edge points and connected edge points in the edge point cloud are traversed, thereby obtaining a contour line set. All strong edge points and connected edge points in each contour line are projected onto a plane. From the leftmost projection point, new projection points are sequentially added according to the first-axis-second-axis rule. If the cross product of three consecutive projection points is less than or equal to 0, the middle projection point is deleted. Until all projection points are traversed, the lower convex hull of each contour line is obtained. From the rightmost projection point, the upper convex hull of each contour line is obtained in the same manner. The convex hulls of each contour line are combined to construct a regular contour. Further, the contour area of the regular contour is calculated by using an image contour detection algorithm. A minimum circumscribed area is determined by using a rotating rectangle fitting algorithm. The rectangular degree of the regular contour is obtained by dividing the contour area by the minimum circumscribed area. The height between the coordinates of the highest point and the lowest point in the regular contour is calculated. When and only when the rectangular degree of the regular contour is greater than a rectangular degree threshold and the height error between the height of the regular contour and the height of a known size of a box is within an error threshold, it is determined that the box exists. Otherwise, it is determined that the box does not exist.

[0020] Further, the contour area of the regular contour is calculated by using an image contour detection algorithm. A minimum circumscribed area is determined by using a rotating rectangle fitting algorithm. The rectangular degree of the regular contour is obtained by dividing the contour area by the minimum circumscribed area. The height between the coordinates of the highest point and the lowest point in the regular contour is calculated. When and only when the rectangular degree of the regular contour is greater than a rectangular degree threshold and the height error between the height of the regular contour and the height of a known size of a box is within an error threshold, it is determined that the box exists. Otherwise, it is determined that the box does not exist.

[0021] ​​​​Furthermore, the identification and decision module includes a visual recognition unit, a comprehensive authentication unit, and a response processing unit;

[0022] The visual recognition unit senses the garbage truck through an electromagnetic coil and takes a picture of the original vehicle. The original vehicle image was extracted using an atmospheric light algorithm. The dark channel in the image is optimized based on transmittance to obtain the defogging vehicle image. Super-resolution reconstruction model is used to analyze defogging vehicle images. High-resolution features are obtained by performing convolution and deconvolution. The garbage truck model and license plate were identified by using feature database comparison and image segmentation model.

[0023] The integrated authentication unit reads the tag ID and VIN code and encrypts them with a randomly generated session key. It then encrypts the session key using the server's public key, packages it, and sends it to the authorization server. The server receives the encrypted data, decrypts it to obtain the garbage truck model and vehicle license plate, and compares it with the garbage truck model and vehicle license plate identified by the visual recognition unit. If they match, a bin inspection command is sent; otherwise, no action is taken. The encrypted data is obtained by the authorization server when querying the whitelist and matching the tag ID and VIN code using the session key and the encrypted garbage truck model and vehicle license plate.

[0024] The processing unit receives a comparison command and determines whether a waiting command exists. If no waiting command exists, it waits; if a waiting command exists, it deletes the waiting command and determines the remaining volume of the enclosure using infrared ranging. and with volume threshold Compare; if it is less than or equal to the volume threshold Send a transfer instruction if the volume exceeds the threshold. By using millimeter-wave radar array scanning, the radial distance of each echo signal is calculated based on the radar ranging principle. Combined with the installation angle of each radar Constructing a projected data distribution with the amplitude of each echo signal The attenuation coefficient distribution is updated by employing a hierarchical iterative algorithm. Grayscale images are generated based on grayscale conversion and layered 3D watershed segmentation is performed. Morphological optimization and geometric calculations are then used to obtain the volume of waste carried. Compare the remaining volume of the box With the volume of garbage carried If the remaining volume of the box Greater than or equal to the volume of the waste carried Send a full unload command if the remaining volume of the enclosure is... Smaller than the volume of the waste it carries Send the unloading command.

[0025] Specifically, atmospheric light algorithms determine the original vehicle image. Mid-pixel coordinates local neighborhood , local neighborhood The minimum pixel value among the three inner channels is used as the pixel coordinate. Dark channel pixel values And calculate pixel coordinates transmittance Traverse the original vehicle map The original vehicle image is obtained by taking the grayscale value of each pixel coordinate, removing the top 5% of pixels with the highest grayscale value, and then averaging the grayscale values ​​of the remaining pixels. Atmospheric light value Based on the atmospheric scattering model, atmospheric light values ​​are... minus transmittance The product of these factors represents the brightness deviation caused by fog, and the original vehicle image is used as the basis for this. Subtract the brightness deviation caused by fog and divide by the transmittance Obtain the defogging vehicle image .

[0026] Specifically, the super-resolution reconstruction model includes a deep residual network and a sampling-to-restore network. The deep residual network extracts defogging vehicle images through convolution. Shallow features in Sampling and restoration networks for shallow features Pixel convolution and deconvolution are performed sequentially to generate high-resolution features. Calculate high-definition features The garbage truck model is determined based on the similarity of vehicle features with each vehicle in the vehicle feature database, using maximum similarity matching. An image segmentation model is employed, and the dehazed vehicle images are progressively extracted and compressed through continuous downsampling. From the feature dimensions, we obtain abstract features. By continuously upsampling to restore the feature dimensions and locate the license plate area, the license plate area mask is obtained. The OCR engine is then called to recognize the license plate area mask to obtain the vehicle license plate.

[0027] Furthermore, the infrared ranging method obtains the measured distance using an array of infrared ranging sensors and organizes it into a ranging matrix. Using the least squares method, a polynomial fitting is performed based on the ranging matrix to determine the representation function of the waste surface. The remaining volume of the container is calculated by integrating the representation function of the waste surface based on the coverage area of ​​the top surface of the container. The top surface of the container is perpendicular to the direction of garbage compression and does not come into contact with the garbage when the container is not full.

[0028] Specifically, a millimeter-wave radar array is used to transmit millimeter-wave signals from different angles and record the transmission timestamps. Each echo signal is received and the reception timestamp, amplitude, and phase are recorded. Based on the radar ranging principle, the radial distance of each echo is determined. It is equal to half the product of the speed of light and the signal transmission time, where the signal transmission time of each echo signal is equal to the receiving timestamp minus the transmitting timestamp, depending on the installation angle of the radar corresponding to each echo signal. The amplitude of each echo signal is determined according to the installation angle. and radial distance Arranged as projected data distribution .

[0029] Furthermore, the hierarchical iterative algorithm includes the following steps:

[0030] The garbage truck's cavity is evenly divided along the garbage compression direction. Layer and initialize the first Subdistribution of layer attenuation coefficient , ;

[0031] Calculation of installation angle based on ensemble ray tracing method and radial distance The line integral weight function below ;

[0032] In the In this iteration, the distribution of the projected data is calculated. Expected likelihood value ;

[0033] Using Poisson likelihood function Description of the The distribution of attenuation coefficients in the next iteration The distribution of the observed projection data The probability of;

[0034] Based on Poisson likelihood function The logarithmic form of the th The second iteration Subdistribution of layer attenuation coefficient The first and second partial derivatives are updated to obtain the first... The second iteration Subdistribution of layer attenuation coefficient ;

[0035] Repeat the iteration until two consecutive iterations are complete. The iteration stops when the average error of the attenuation coefficient subdistribution of the layer is less than the convergence threshold, thus obtaining the attenuation coefficient distribution. .

[0036] Specifically, the attenuation coefficient distribution is divided into a plurality of layer attenuation coefficient sub-distributions converted into a gray scale image by being normalized to a range of 0 to 1 and multiplied by a maximum gray scale value, the gray scale image is divided into a plurality of layer gray scale sub-images according to the layer attenuation coefficient sub-distributions, the gradient amplitudes of each layer gray scale sub-image are calculated and compared with the amplitude threshold value of each layer, and the positions greater than the amplitude threshold value of each layer are determined as ridge lines of the watershed, wherein the amplitude threshold value of the first layer is the amplitude initial value multiplied by the number of layers The layer gray scale sub-images are divided into a plurality of layer gray scale sub-images according to the layer attenuation coefficient sub-distributions The amplitude threshold value of each layer is the amplitude initial value multiplied by the number of layers The amplitude threshold value of each layer is the amplitude initial value multiplied by the number of layers The ratio of the number of layers to the total number of layers, for each region segmented by the ridge line of the watershed, morphological dilation is performed using a spherical structure element, and a boundary matching algorithm is used to ensure interlayer connection, regions with a gray scale value greater than a gray scale threshold value in the gray scale image are determined as garbage, and the volume of the garbage is calculated by geometric knowledge .

[0037] Compared with the prior art, the model and license of the garbage truck and the transfer vehicle are identified and verified through visual recognition and NFC identification technology, the access permission of the garbage dumping and transfer link is accurately controlled, meanwhile, a data interaction system among the box detection, identification judgment, compression execution and transfer processing modules is constructed, real-time data is obtained by using infrared ranging and radar scanning technologies, and dynamic decision is realized based on an algorithm model. The instructions are sequentially transmitted among the modules, forming a closed-loop control logic from detection to transfer, and a multi-modal identity verification mechanism prevents illegal dumping behavior and avoids safety hazards and disorder in the garbage disposal link; the closed-loop control realizes efficient cooperation of each link of the garbage compression processing, eliminates data islands and decision layers, makes vehicle scheduling more scientific and reasonable, reduces waiting time and resource waste, and finally realizes intelligent, standardized and efficient operation of the garbage compression processing flow. BRIEF DESCRIPTION OF DRAWINGS

[0038] Figure 1 Fig. 1 is a schematic diagram of a garbage compression intelligent control system;

[0039] Figure 2 Fig. 4 is a processing flowchart of the processing feedback unit;

[0040] Figure 3 Fig. 5 is a processing flowchart of the response processing unit;

[0041] Figure 4 Fig. 6 is a flowchart of the hierarchical iteration algorithm. DETAILED DESCRIPTION

[0042] The application will be further described in detail below in combination with the drawings and examples.

[0043] Example 1

[0044] As Figure 1 ​As shown in the figure, a specific embodiment of the present invention discloses an intelligent control system for waste compression, including a container detection module, an identification and decision module, a compression execution module, and a transfer and processing module;

[0045] The enclosure detection module scans and acquires the point cloud inside the cavity when the system starts up or receives an enclosure inspection command. It performs preprocessing to preserve the contour, determines the regular contour through edge detection, contour tracking and convex hull calculation, calculates geometric features and performs rule verification, and adaptively selects to send a transfer instruction or a feedback waiting instruction or a feedback comparison instruction.

[0046] The identification and judgment module uses visual recognition technology to identify the garbage truck model and license plate, combined with NFC recognition for multi-dimensional identity verification. Upon successful identity verification, it sends a container inspection command to the container detection module. If a waiting command is received, no action is taken. If a comparison command is received, it checks whether a waiting command exists. If not, it waits; if a waiting command exists, it deletes the waiting command and determines the remaining volume of the container using infrared ranging. It then decides whether to send transfer instructions or use a millimeter-wave radar array for scanning, and constructs a projection data distribution based on radar ranging principles and echo signals. The attenuation coefficient distribution is then updated using a hierarchical iterative algorithm. The volume of the waste carried was obtained based on grayscale conversion and three-dimensional topological segmentation. Compare the remaining volume of the box With the volume of garbage carried The decision is made to send either a full unloading command or a partial unloading command, wherein, and These represent the installation angle of the radar that generates the echo signal and the radial distance of the echo signal, respectively. Represents spatial coordinates;

[0047] The compression execution module receives the full unloading command, opens the feed inlet to wait for the garbage truck to dump, and continuously monitors the remaining volume of the container using infrared ranging. Changes will continue until the remaining volume of the enclosure remains within the preset time period. Initiate compression when there is no change, and determine the remaining volume of the chamber again using infrared ranging after compression is complete. and with volume threshold Comparison, if the remaining volume of the box Less than or equal to the volume threshold If the container is considered full, a transfer instruction will be sent. If the container has remaining volume... Greater than the volume threshold It does not take any action and waits for a new garbage truck to arrive. Upon receiving the unloading instruction, it opens the feed inlet and waits for the garbage truck to dump the waste. It also continuously monitors the remaining volume of the container using infrared ranging. until the remaining volume of the box Equal to the volume threshold When the box is full, the inlet is immediately closed and a stop dumping prompt is sent to the garbage truck, the compression is started and the comparison instruction is fed back after the compression is completed;

[0048] The transfer processing module receives the transfer instruction and sends a transfer notification to the transfer vehicle. The carrying platform is raised to a specified height. The empty box is determined to be in place through the limit switch state and weight sensing. When the limit switch is in place and the error between the platform lifting weight and the standard empty box weight within a specified time is always within the weight error threshold, the empty box is determined to be in place. The transfer vehicle model and vehicle license plate are identified by visual recognition technology. Multi-dimensional identity verification is performed in combination with NFC identification. After successful identity verification, the carrying platform is lowered to the original position and a comparison instruction is fed back to the identification decision module. The visual recognition technology, identity verification method and identification decision module are consistent and will not be described in detail.

[0049] Further, the box detection module includes a detection triggering unit, a detection analysis unit and a processing feedback unit.

[0050] The detection triggering unit controls the laser scanner to perform line-by-line laser scanning on the cavity when the system is started or receives a box detection instruction, and obtains the cavity point cloud .

[0051] The detection analysis unit performs downsampling and filtering on the cavity point cloud to retain the outline, identifies the edge points based on edge detection, and determines the regular outline by contour tracking and convex hull calculation. The rectangularity and height are calculated by geometric shape feature analysis and rule verification is performed.

[0052] As shown in Figure 2 , the processing feedback unit makes adaptive decisions based on the detection triggering time and the rule verification result. If the detection triggering time is system startup and the rule verification result is no box, a transfer instruction is sent to the transfer processing module. If the detection triggering time is system startup and the rule verification result is a box, no processing is performed and the garbage truck is waited. If the detection triggering time is receiving a box detection instruction and the rule verification result is no box, the transfer vehicle has been notified to send an empty box but the transfer vehicle has not arrived. A wait instruction is fed back to the identification decision module to wait for the arrival of the empty box. If the detection triggering time is receiving a box detection instruction and the rule verification result is a box, a comparison instruction is fed back to the identification decision module to make a decision on the garbage unloading strategy.

[0053] Specifically, after the detection analysis unit obtains the cavity point cloud , the three-dimensional space is divided into cubic voxels with a fixed edge length and sequentially indexed, wherein the edge length is pre-set according to the box size. Based on the cavity point cloud The coordinates of each intracavity point are determined to be located in a cubic voxel, and a corresponding index number is assigned to the intracavity point coordinates. The centroid coordinates of intracavity point coordinates with the same index number are calculated and used to replace the corresponding intracavity point coordinates within the cubic voxel to achieve redundant point merging. Downsampling is then used to generate a centroid point cloud. While preserving key geometric features, this reduces the complexity of subsequent processing and improves computational efficiency.

[0054] Furthermore, bilateral filtering is used to process the centroid point cloud. To remove noise, based on spatial distance and kernel bandwidth Search for the centroid cloud The Middle Coordinates of the centroid point neighborhood Calculate the neighborhood covariance matrix The details are as follows:

[0055] ,

[0056] in, For the neighborhood Inner Coordinates of the centroid point Represents the transpose of a matrix or vector. , For the neighborhood The total number of points in the neighborhood covariance matrix Perform eigenvalue decomposition and select the eigenvector corresponding to the smallest eigenvalue as the first eigenvalue. Coordinates of the centroid point normal vector Based on neighborhood Inner Coordinates of the centroid point With the Coordinates of the centroid point Calculate the distance weights using the spatial distance and the normal vector respectively. and similarity weight The details are as follows:

[0057] ,

[0058] ,

[0059] in, This represents an exponential function with the natural constant as its base. For the first Coordinates of the centroid point The normal vector, For feature similarity kernel bandwidth, based on neighborhood each centroid point coordinate in the neighborhood is weighted and summed up and divided by the corresponding weight product, to obtain the th denoised point coordinate , as follows:

[0060] ,

[0061] wherein, each centroid point coordinate in the centroid point cloud is traversed and bilateral filtered, effectively weakening random noise caused by measurement errors and environmental interference, because the similarity weight of edge points is low, the characteristics of edge points are completely retained during bilateral filtering, avoiding the edge blur problem of traditional filtering, to generate a denoised point cloud .

[0062] Further, the detection and analysis unit calculates the gradient of the th denoised point coordinate in the denoised point cloud , calculates the gradient amplitude and gradient direction of the th denoised point coordinate , compares the gradient amplitude of the th denoised point coordinate with the gradient amplitudes of adjacent denoised point coordinates in the gradient direction, if the gradient amplitude of the th denoised point coordinate is not a local maximum, the gradient amplitude of the th denoised point coordinate is suppressed to 0, if the gradient amplitude of the th denoised point coordinate is a local maximum, the gradient amplitude of the th denoised point coordinate is retained, a high threshold and a low threshold are set, denoised coordinate points with gradient amplitudes greater than the high threshold are marked as strong edge points, denoised coordinate points with gradient amplitudes between the high threshold and the low threshold and connected to the strong edge points are marked as connected edge points, and the gradient amplitudes of the remaining denoised coordinate points are all suppressed to 0, to obtain an edge point cloud .

[0063] Further, based on contour tracking, an edge point cloud ​​​​If an untracked strong edge point is found, neighboring edge points are searched sequentially in a clockwise direction based on neighborhood search until the initial strong edge point is found or no new connected edge point can be found, resulting in a continuous contour line. This process is repeated until the edge point cloud is traversed. Given all strong edge points and connected edge points in the contour, obtain the contour set. Project the set of points formed by all strong edge points and connected edge points in each contour line onto the map. The plane, starting from the leftmost projection point, is based on the prior... Rear of axle The rules of the axis are to add new projection points in sequence. If the cross product of three consecutive projection points is less than or equal to 0, the middle projection point is deleted. This process is repeated until all projection points are traversed to obtain the lower convex hull of each contour line. Starting from the rightmost projection point, the upper convex hull of each contour line is obtained in the same way. The convex hulls of each contour line are then merged to form a regular contour. The convex hulls retain the overall convex shape of the contour line and filter out contour fluctuations caused by local noise or small depressions on the object surface.

[0064] Furthermore, the contour area of ​​the regular contour is calculated using an image contour detection algorithm, and the minimum bounding area of ​​the minimum bounding rectangle is determined using a rotating rectangle fitting algorithm. The rectangularity of the regular contour is obtained by dividing the contour area by the minimum bounding area. The image contour detection algorithm and the rotating rectangle fitting algorithm are existing algorithms and will not be elaborated on. The rectangularity is used to measure whether the regular contour is a regular shape. The closer the rectangularity is to 1, the more regular the shape of the regular contour. The height between the coordinates of the highest point and the coordinates of the lowest point in the regular contour is calculated. A box is determined to exist if and only if the rectangularity of the regular contour is greater than the rectangularity threshold and the height error between the height of the regular contour and the height of the box of known size is within the error threshold. Otherwise, no box is determined to exist.

[0065] Furthermore, the identification and decision module includes a visual recognition unit, a comprehensive authentication unit, and a response processing unit;

[0066] The visual recognition unit senses the impact of garbage trucks by using electromagnetic coils deployed under the lanes, and sends a wake-up signal to the camera via the GIPO bus to control the capture of the original vehicle image. The original vehicle image was extracted using an atmospheric light algorithm. The dark channel in the image is optimized based on transmittance to obtain the defogging vehicle image. Super-resolution reconstruction model is used to analyze defogging vehicle images. High-resolution features are obtained by performing convolution and deconvolution. The garbage truck model and license plate were identified by using feature database comparison and image segmentation model.

[0067] The integrated authentication unit polls the tags on the vehicle based on the NFC-A protocol, reads the tag ID and VIN code, randomly generates a session key to encrypt the tag ID and VIN code, and then packages and sends the session key to the authorization server after encrypting it with the server's public key. The unit receives the encrypted data from the server and decrypts it to obtain the garbage truck model and vehicle license plate. It compares and verifies the garbage truck model and vehicle license plate identified by the visual recognition unit. If they are exactly the same, a container inspection command is sent to the container detection module. If they are different, no processing is performed. The encrypted data is obtained by the authorization server querying the whitelist in the MySQL database and successfully matching the tag ID and VIN code, and then encrypting the garbage truck model and vehicle license plate using the session key.

[0068] like Figure 3 As shown, the processing unit receives a comparison command and determines whether a waiting command exists. If no waiting command exists, it indicates that no garbage truck is waiting to dump, so no processing is performed. If a waiting command exists, it indicates that a garbage truck is waiting to dump, so the waiting command is deleted, and the remaining volume of the container is determined by infrared ranging. If the remaining volume of the box Less than or equal to the volume threshold If the container is considered full, a transfer instruction will be sent. If the remaining volume of the container is... Greater than the volume threshold The system uses a millimeter-wave radar array for scanning and calculates the radial distance of each echo signal based on radar ranging principles. Combined with the installation angle of each radar Constructing a projected data distribution with the amplitude of each echo signal The attenuation coefficient distribution is updated by employing a hierarchical iterative algorithm. Grayscale images are generated based on grayscale conversion and layered 3D watershed segmentation is performed to distinguish waste regions of different densities. Morphological optimization and geometric calculations are then used to obtain the volume of waste carried. Compare the remaining volume of the box With the volume of garbage carried If the remaining volume of the box Greater than or equal to the volume of the waste carried Send a full unload command if the remaining volume of the enclosure is... Smaller than the volume of the waste it carries Send the unloading command.

[0069] Specifically, atmospheric light algorithms determine the original vehicle image. Mid-pixel coordinates local neighborhood , pixel coordinates The pixel values ​​in the R, G, and B channels are taken as local neighborhood values. The minimum value among the three inner channels is used to generate pixel coordinates. Dark channel pixel values Among them, the dark channel pixel value Highlight the original vehicle image Mid-pixel coordinates The darkest local information reflects the trend of fog influence; pixel coordinates are calculated. transmittance transmittance Reflecting the proportion of light remaining after it penetrates the fog, traversing the original vehicle image. The original vehicle image is obtained by taking the grayscale value of each pixel coordinate, removing the top 5% of pixels with the highest grayscale value, and then averaging the grayscale values ​​of the remaining pixels. Atmospheric light value Based on the atmospheric scattering model, the original vehicle image Subtracting the brightness deviation caused by fog and restoring the true brightness of light after passing through the fog, we obtain the defogging vehicle image. The details are as follows:

[0070] ,

[0071] in, and Images of vehicles undergoing defogging and the original vehicle image Mid-pixel coordinates The pixel value.

[0072] Specifically, the super-resolution reconstruction model includes a deep residual network and a sampling and restoration network. The deep residual network consists of 16 layers of residual blocks, which extract defogging vehicle images through convolution. Shallow features in The sampling and restoration network utilizes pixel convolution to process shallow features. Upsampling is performed, and the data is then deconvolved to restore the features to their original state. Generate high-resolution features in the same dimension Calculate high-definition features The garbage truck model is determined based on the similarity of vehicle features with each vehicle in the vehicle feature database, using maximum similarity matching. An image segmentation model based on an autoencoder is used to progressively extract and compress dehazed vehicle images through five consecutive downsampling layers. From the feature dimensions, we obtain abstract features. By upsampling five consecutive layers to restore the feature dimensions and locate the license plate area, the license plate area mask is obtained. The OCR engine is then called to recognize the license plate area mask to obtain the vehicle license plate.

[0073] Further, the infrared ranging method vertically emits infrared light to the garbage in the box through infrared ranging sensors arranged in an array, records the measurement distances of different infrared ranging sensors and arranges the measurement distances into a ranging matrix according to the spatial distribution of the infrared ranging sensors, wherein the measurement distance refers to the distance from the reflection point of the infrared light on the garbage surface to the top surface of the box, adopts the least square method, performs polynomial fitting on the ranging matrix, determines the expression function of the garbage surface about the spatial coordinates, performs micro-element volume integration on the expression function of the garbage surface about the spatial coordinates according to the coverage range of the top surface of the box in the spatial coordinate system, and calculates the remaining volume of the box , wherein the top surface of the box is defined as the surface of the box that is perpendicular to the garbage compression direction and does not contact the garbage when the garbage is not full.

[0074] Specifically, the millimeter wave radar array is distributed in a ring shape, emits millimeter wave signals to the garbage truck cavity from different angles and records the emission time stamps, receives each echo signal after passing through the cavity, records the reception time stamp, amplitude and phase of each echo signal, based on the radar ranging principle, the radial distance of each echo is equal to half of the product of the speed of light and the signal transmission time length, wherein the signal transmission time length of each echo signal is equal to the reception time stamp minus the emission time stamp of each echo signal, according to the installation angle of the radar corresponding to each echo signal , the amplitude of each echo signal is projected into a two-dimensional angle-distance coordinate system according to the installation angle and the radial distance , to form a projection data distribution , the projection data distribution reflects the distribution of the amplitude at the installation angle and the radial distance .

[0075] As shown in Figure 4 , further, the hierarchical iterative algorithm includes the following steps:

[0076] Under the assumption of uniform layering, the garbage truck cavity is uniformly divided into layers along the garbage compression direction, and the attenuation coefficient sub-distribution of the first layer is initialized as a fixed constant value ;

[0077] Based on the set ray tracing method, the line integral weight function can be calculated when the installation angle and the radial distance are given, and the line integral weight function is essentially the contribution degree of the spatial coordinates on the straight line with an angle of to the ray, which reflects the contribution degree of the spatial coordinates​ The collective ray tracing method is an existing algorithm and will not be elaborated here.

[0078] In the first iteration, the expected likelihood value of the projection data distribution is calculated, as follows:

[0079] ,

[0080] wherein is the attenuation coefficient distribution of the first iteration; The Poisson likelihood function

[0081] is used to describe the probability of observing the projection data distribution in the first iteration, as follows:

[0082] ,

[0083] wherein denotes the product of the function for all feasible combinations of the installation angle and the radial distance , denotes the factorial of the amplitude for the installation angle and the radial distance .

[0084] The attenuation coefficient sub-distribution of the first iteration of the layer is updated based on the first-order and second-order partial derivatives of the logarithmic form of the Poisson likelihood function with respect to the attenuation coefficient sub-distribution of the layer of the first iteration, as follows:

[0085] ,

[0086] wherein denotes the spatial coordinates in the layer;

[0087] The iteration is repeated until the average error of the attenuation coefficient sub-distributions of the layers of the adjacent two iterations is less than the convergence threshold, and the iteration is stopped to obtain the final attenuation coefficient distribution . The attenuation coefficient distribution ​​​​​​​​​This reflects the spatial distribution of garbage density within the garbage truck's cavity, providing effective data support for subsequent volume calculations.

[0088] Specifically, the attenuation coefficient distribution By normalizing the image to the range of 0 to 1 using min-max normalization and multiplying it by the maximum grayscale value of 255, a grayscale image is generated. If this grayscale image is considered as a 3D terrain slice, then the grayscale image also includes... Layer grayscale subimage, where the first The grayscale sub-image of the layer is composed of the first Sub-distribution of layer attenuation coefficient The process involves transformation and generation, calculating the gradient magnitude of each grayscale sub-image, comparing it with a magnitude threshold for each layer, and identifying positions where the gradient magnitude exceeds the threshold as watershed ridges. Specifically, the ridge... The amplitude threshold of a layer is the initial amplitude value multiplied by the number of layers. With total number of floors The ratio is used to perform morphological dilation of the regions segmented by each watershed ridge line, and a boundary matching algorithm is used to ensure interlayer connectivity. Regions with gray values ​​greater than the gray threshold in the grayscale image are identified as garbage, and the volume of garbage carried is obtained through geometric calculation. Among them, morphological dilation and boundary matching algorithms are existing technologies and will not be elaborated on further.

[0089] The application discloses a garbage compression intelligent control system, which comprises a box detection module, an identification and decision module, a compression execution module and a transfer processing module; the box detection module acquires and pre-processes the point cloud in the cavity when the system is started or receives a box detection instruction, determines a regular contour through edge detection, contour tracking and convex hull calculation, carries out rule verification based on geometric shape features to accurately determine whether the box exists, and adaptively decides to send a transfer instruction or feedback a waiting instruction or a comparison instruction; the identification and decision module identifies the garbage truck model and vehicle license plate through visual recognition technology, carries out identity verification in combination with NFC identification to avoid random dumping, sends a box detection instruction after the identity verification succeeds, waits when receiving a waiting instruction, and selects to wait or delete the waiting instruction based on whether the waiting instruction exists, determines the remaining volume of the box through an infrared distance measurement method, decides to send a transfer instruction or carries out radar scanning, constructs projection data distribution based on the radar ranging principle and echo signals, and inversely obtains the attenuation coefficient distribution by using a hierarchical iteration algorithm, obtains the garbage volume carried by the box through gray scale conversion and three-dimensional topological segmentation, compares the remaining volume of the box with the garbage volume carried by the box to dynamically decide to send a full unloading instruction or a partial unloading instruction, and realizes dynamic unloading judgment based on the remaining volume of the box and the garbage volume; the compression execution module receives the full unloading instruction, monitors the change of the remaining volume of the box through the infrared distance measurement method, carries out compression when the remaining volume of the box does not change within a preset time length, determines the remaining volume of the box again after compression and compares the remaining volume with a volume threshold to decide to send a transfer instruction or wait, receives the partial unloading instruction, monitors the remaining volume of the box through the infrared distance measurement method, sends a stop dumping prompt and carries out compression when the remaining volume of the box is equal to the volume threshold, and feeds back the comparison instruction after compression; the transfer processing module receives the transfer instruction and sends a transfer notice, lifts the platform and determines whether the empty box is in place through a limit switch and a weight sensing, identifies the transfer truck model and vehicle license plate through visual recognition technology when the empty box is in place, carries out identity verification in combination with NFC identification, lowers the platform and feeds back the comparison instruction after the identity verification succeeds, provides a box changing and unloading integrated closed-loop control with multi-modal identity verification and multi-module data interaction, and realizes efficient garbage compression processing.

[0090] The above only describes the preferred embodiments of the application, and the protection scope of the application is not limited to the above-described embodiments. Any technical solution falling within the concept of the application shall fall within the protection scope of the application. It should be noted that, for ordinary technical personnel in the technical field, some improvements and refinements without departing from the principles of the application shall also be considered as falling within the protection scope of the application.

Claims

1. A garbage compression intelligent control system, characterized in that, The box detection module, the identification decision module, the compression execution module and the transfer processing module are included. The box detection module acquires and pre-processes the cavity point cloud when the system starts or receives the box detection instruction, determines the regular contour through edge detection, contour tracking and convex hull calculation, and performs rule verification based on geometric shape features to select to send the transfer instruction or feedback the waiting instruction or feedback the comparison instruction. The identification decision module identifies the garbage truck model and vehicle license plate and performs identity verification in combination with NFC identification, sends the box detection instruction after verification success, waits when receiving the waiting instruction, and selects to wait or delete the waiting instruction and measures the remaining volume of the box based on whether the waiting instruction exists, decides to send the transfer instruction or measures the carried garbage volume, and compares the remaining volume of the box with the carried garbage volume to decide to send the full unloading instruction or the partial unloading instruction. The compression execution module receives the full unloading instruction, monitors the change of the remaining volume of the box, and performs compression when the remaining volume of the box does not change within a preset time length, determines the remaining volume of the box again after compression and compares it with the volume threshold to decide to send the transfer instruction or wait, receives the partial unloading instruction, monitors the remaining volume of the box, and sends the stop dumping prompt and performs compression when the remaining volume of the box is equal to the volume threshold, and feedbacks the comparison instruction after compression. The transfer processing module receives the transfer instruction and sends the transfer notification, lifts the platform and determines whether the empty box is in place through the limit switch and the weight induction, identifies the transfer truck model and vehicle license plate after the empty box is in place, performs identity verification in combination with NFC identification, lowers the platform after verification success and feedbacks the comparison instruction.

2. The intelligent control system for garbage compression according to claim 1, wherein, The box detection module includes a detection trigger unit, a detection analysis unit and a processing feedback unit. The detection trigger unit performs laser scanning on the cavity to acquire the cavity point cloud when the system starts or receives the box detection instruction. The detection analysis unit performs downsampling and filtering on the cavity point cloud, identifies the edge points based on edge detection, and determines the regular contour by contour tracking and convex hull calculation, calculates the rectangularity and height and performs rule verification. The processing feedback unit sends the transfer instruction when the detection trigger occasion is system startup and the rule verification result is no box, waits when the detection trigger occasion is system startup and the rule verification result is a box, feedbacks the waiting instruction when the detection trigger occasion is receiving the box detection instruction and the rule verification result is no box, and feedbacks the comparison instruction when the detection trigger occasion is receiving the box detection instruction and the rule verification result is a box.

3. The intelligent control system for garbage compression as claimed in claim 2, wherein, The filtering adopts bilateral filtering, searches a neighborhood of a first centroid point coordinate in a centroid point cloud, calculates a covariance matrix, performs eigenvalue decomposition on the covariance matrix, determines a normal vector of the first centroid point coordinate, respectively calculates a distance weight and a similarity weight between each centroid point coordinate in the neighborhood and the first centroid point coordinate, performs weighted summation on a corresponding centroid point coordinate and divides the corresponding centroid point coordinate by a corresponding weight product, to obtain a first denoised point coordinate, and performs bilateral filtering on each centroid point coordinate in the centroid point cloud to generate a denoised point cloud, wherein the centroid point cloud is a down-sampling result of the intracavity point cloud. ​ 4. The intelligent control system for garbage compression as claimed in claim 2, wherein, Based on contour tracking, selecting strong edge points in the edge point cloud which are not tracked, searching adjacent edge points along the clockwise direction until returning to the starting strong edge point or stopping when a new connected edge point cannot be searched, obtaining a continuous contour line, repeating the process until all strong edge points and connected edge points are traversed, obtaining a contour line set, projecting the strong edge points and connected edge points in each contour line to a plane, starting from the leftmost projection point, adding new projection points in turn according to a preset rule, if the cross product of three continuous projection points is less than or equal to 0, deleting the middle projection point, until all projection points are traversed, obtaining the lower convex hull of each contour line, starting from the rightmost projection point, obtaining the upper convex hull of each contour line in the same way, merging to obtain the convex hull of each contour line, and combining the convex hulls of each contour line to construct a regular contour, wherein the edge point cloud, the strong edge points and the connected edge points are obtained by edge detection.

5. The intelligent control system for refuse compression as claimed in claim 1, wherein, The recognition decision module comprises a visual recognition unit; The visual recognition unit inducts the garbage truck through an electromagnetic coil and takes an original vehicle image, extracts a dark channel in the original vehicle image by using an atmospheric light algorithm and obtains a defogged vehicle image according to transmittance optimization, convolves and deconvolves the defogged vehicle image by using a super-resolution reconstruction model to obtain high-definition features, and identifies the garbage truck model and vehicle license plate by using feature library comparison and an image segmentation model.

6. The intelligent control system for refuse compression as claimed in claim 5, wherein, The atmospheric light algorithm determines the local neighborhood of a pixel coordinate in the original vehicle image, takes the minimum pixel value in the local neighborhood as the dark channel pixel value of the pixel coordinate and calculates the transmittance of the pixel coordinate, traverses the gray scale of each pixel coordinate in the original vehicle image, removes 5% of the pixel coordinates with the largest gray scale and takes the gray scale average of the remaining pixel coordinates to obtain the atmospheric light value, based on an atmospheric scattering model, takes the product of the atmospheric light value and 1 minus the transmittance as the brightness deviation caused by fog, subtracts the brightness deviation caused by fog from the original vehicle image and divides by the transmittance to obtain the defogged vehicle image.

7. The intelligent control system for refuse compression as claimed in claim 1, wherein, The recognition decision module further comprises a coping processing unit; The coping processing unit receives a comparison instruction, then selects waiting or deletes the waiting instruction based on whether the waiting instruction exists, determines the remaining volume of the box by infrared distance measurement and compares it with the volume threshold, if it is less than or equal to the volume threshold, sends a transfer instruction, if it is greater than the volume threshold, scans by using a millimeter wave radar array, calculates the radial distance of each echo signal based on the radar ranging principle, constructs the projection data distribution by combining the installation angle of each radar and the amplitude of each echo signal, updates the attenuation coefficient distribution by using a hierarchical iterative algorithm, generates a gray scale image based on gray scale conversion and performs hierarchical three-dimensional watershed segmentation, obtains the garbage carrying volume by using morphological optimization and geometric calculation, compares the remaining volume of the box with the garbage carrying volume, if the remaining volume of the box is greater than or equal to the garbage carrying volume, sends a full unloading instruction, and if the remaining volume of the box is less than the garbage carrying volume, sends a partial unloading instruction.

8. The intelligent control system for refuse compression as claimed in claim 7, wherein, The millimeter wave radar array emits millimeter wave signals from different angles and records the emission time stamp, receives each echo signal and records the receiving time stamp, amplitude and phase, and based on the radar ranging principle, the radial distance of each echo is equal to half of the product of the speed of light and the signal transmission time length, wherein the signal transmission time length of each echo signal is equal to the receiving time stamp minus the emission time stamp, and the amplitude of each echo signal is arranged as a projection data distribution according to the installation angle and the radial distance according to the installation angle of the radar corresponding to each echo signal.

9. The intelligent control system for refuse compression as claimed in claim 7, wherein, The hierarchical iterative algorithm comprises the following steps: The cavity of the garbage truck is evenly divided into layers along the garbage compression direction and the attenuation coefficient sub-distribution of the first layer is initialized, ; Calculating a line integral weight function under the installation angle and the radial distance based on the set ray tracing method; In the In each iteration, the expected likelihood value of the projected data distribution is calculated; The Poisson likelihood function is used to describe the probability of observing the projection data distribution given the attenuation coefficient distribution of the nth iteration iteration. Based on the logarithmic form of the Poisson likelihood function with respect to the first... The second iteration The first and second partial derivatives of the attenuation coefficient subdistribution of the layer are updated to obtain the first... The second iteration The attenuation coefficient sub-distribution of the layer; Repeat the iteration until two consecutive iterations are complete. The iteration stops when the average error of the attenuation coefficient sub-distribution of the layer is less than the convergence threshold, and the attenuation coefficient sub-distribution is obtained.

10. The intelligent control system for refuse compression as claimed in claim 9, wherein, The attenuation coefficient distribution is normalized to the range of 0 to 1 and multiplied by the maximum gray value to convert into a gray image, and the gray image is divided into The layer attenuation coefficient sub-distribution is divided into The layer gray sub-image is calculated, the gradient amplitude of each layer gray sub-image is compared with the amplitude threshold value of each layer, and the position greater than the amplitude threshold value of each layer is determined as the ridge line of the watershed, wherein the first The amplitude threshold value of each layer is the amplitude initial value multiplied by the number of layers The ratio of the total number of layers For each region segmented by the ridge line of the watershed, a spherical structure element is used for morphological expansion, and a set boundary matching algorithm is used to ensure interlayer connection. The region with a gray value greater than the gray threshold value in the gray image is determined as garbage, and the volume of the garbage is calculated.

Images