Logistics park operation site nonlinear event electronic grid method

Abstract

The invention discloses a logistics park operation site nonlinear event electronic grid method. The method comprises the following steps that S1, a high-definition monitoring camera collects video data of an operation area; s2, transmitting the video data to an artificial intelligence dedicated server; s3, identifying a nonlinear event by the artificial intelligence dedicated server; s4, identifying a non-linear time correlation region; and S5, performing nonlinear event electronic grid virtual isolation. According to the method, an artificial intelligence recognition technology is adopted, nonlinear events are dynamically tracked and recognized, occurrence of the nonlinear events is found as early as possible, the types and parameters of the nonlinear events are evaluated, associated operation areas are determined, cross-area electronic grid virtual isolation is achieved, and the operation efficiency is improved. The negative effects of nonlinear events on other operations of the whole logistics park operation site are reduced.

Description

Technical Field

[0001] The invention belongs to the technical field of logistics, and in particular relates to an electronic grid method for nonlinear events at an operation site of a logistics park. Background Art

[0002] Logistics parks are areas where logistics operations are concentrated. They are the mixing and connection points of various logistics operations and logistics equipment. As a comprehensive logistics service platform, logistics parks have complex operating characteristics, which are mainly reflected in: diversified logistics service requirements. Logistics parks meet the diversified logistics service needs of different types of enterprises, such as warehousing, transportation, loading and unloading, packaging, distribution, etc., requiring logistics parks to have all-round and integrated logistics service capabilities; secondly, logistics parks have complex processes and operate efficiently in various types of logistics operations to ensure smooth, accurate and safe flow of goods and improve logistics efficiency.

[0003] To this end, logistics parks need advanced logistics technology support. For example, logistics parks use advanced automated warehouses, intelligent logistics systems, unmanned delivery, etc. to improve the automation and intelligence of logistics operations; these systems provide logistics parks with powerful information processing capabilities, realizing real-time monitoring and analysis of information such as cargo flow, vehicle scheduling, and inventory management, thereby improving the accuracy of logistics decisions.

[0004] On this basis, logistics parks also have very high requirements for the safety management of operations, ensuring that goods are stored, transported, loaded and unloaded safely, and reducing the risk of failure and loss of goods, personnel and equipment. In the actual operation of logistics parks, in order to avoid unsafe incidents, or to adjust the operation process in time after an unsafe incident occurs, to minimize the negative impact of such incidents on other surrounding operations, thereby improving the overall operation efficiency of the logistics park, it is a problem that is often encountered and must be reasonably solved in the daily operation of the logistics park.

[0005] The so-called nonlinear events at the logistics park operation site refer to the occurrence of unknown, unplanned and unsafe events during the normal operation process of the logistics park, which affects some areas, equipment, personnel, goods and orders in the logistics park, and requires a nonlinear scheduling process to readjust the original planned scheduling process.

[0006] Regarding the handling of nonlinear events at the operation site of the logistics park, in the actual operation site, the higher the level of automation of the logistics park, the lower the probability of nonlinear events. However, once they occur and are not handled in time, the risks and losses they bring will be higher. On the other hand, there are a large number of mixed operation types in the logistics park, such as mixed operation scenarios such as manual, automated equipment, and mechanical equipment. Although such mixed operation scenarios are more flexible, the probability of nonlinear events is also higher. Based on existing technical conditions, it is often adopted to improve the level of on-site management to achieve the coordination of people, vehicles, goods, and orders.

[0007] The invention patent "A method and device for optimizing logistics operations" with announcement number CN109886618B discloses a solution for optimizing logistics operations by using big data technology to obtain data such as delivery addresses associated with user portraits through estimated processing after receiving user orders. This solution has a certain degree of foresight and advance prediction for order processing, which is conducive to resolving the uncertainty caused by nonlinear events in logistics operations.

[0008] The invention with the announcement number CN 111115084B, "A logistics optimization control system and method for maximizing delivery time", discloses that by analyzing the bill of materials data of the logistics unit, a complete set of data is obtained to predict the equipment, and the logistics plan is prioritized to achieve a logistics operation optimization solution that maximizes delivery time. This solution also uses the technical advantages of big data to provide a technical solution to resolve the losses caused by non-linear events in logistics operations, such as waiting and congestion.

[0009] The invention "A Tobacco Logistics Control System" with announcement number CN 108510222B discloses the use of digital technology to coordinate tobacco logistics control, achieve timely, accurate and comprehensive grasp and processing of tobacco logistics information, and improve the efficiency and accuracy of tobacco logistics distribution. This technical solution provides an overall solution for nonlinear events in collaborative logistics operations, but has higher requirements for the coordination of factors such as people, equipment, goods, and vehicles in the logistics system.

[0010] Therefore, from the existing public technology, digitalization, big data and other technologies can be used to make preventive and adaptive adjustments for nonlinear events in logistics operations, but logistics sites are often highly complex. The operators, operating equipment, operating vehicles and other elements are running at high speed. Although the existing digital technology can play an auxiliary and collaborative role, it has high requirements for the digital use ability of operators. When and after nonlinear events in logistics operations occur, there is a lack of corresponding technical means to avoid secondary events and reduce the losses caused by them. Summary of the invention

[0011] In order to solve the above technical problems existing in the prior art, the present invention provides an electronic grid method for nonlinear events at the operation site of a logistics park. The present invention uses artificial intelligence technology to intelligently identify dynamic operation areas at the logistics operation site, and implements the so-called electronic grid technology to virtually isolate multiple logistics operation areas, so as to achieve synchronization and coordination of multiple logistics operations in time and space, and reduce the adverse effects caused by nonlinear events in logistics operations.

[0012] The technical solution adopted by the present invention is:

[0013] An electronic grid method for nonlinear events at a logistics park operation site, characterized in that the method comprises the following steps:

[0014] Step S1: A high-definition surveillance camera collects video data of the operation area;

[0015] Step S2: The video data is transmitted to a dedicated artificial intelligence server;

[0016] Step S3: The artificial intelligence dedicated server identifies nonlinear events;

[0017] Step S4: nonlinear time-related region identification;

[0018] Step S5: Virtual isolation of nonlinear event electronic grid.

[0019] Furthermore, in step S3, when a suspected nonlinear event is identified, an electronic grid is set up for virtually isolating the associated operating area affected by the nonlinear event. The associated operating area may be in the same operating area A as the nonlinear event, or may be across regions in an adjacent operating area B. The nonlinear event includes at least the following operating abnormalities: operator abnormality, automatic equipment failure, and operating cargo abnormality.

[0020] Furthermore, in step S3, the historical nonlinear events and event types of the surveillance video data are annotated, and an artificial intelligence model is trained to realize intelligent recognition of nonlinear events. The specific steps include two stages, namely:

[0021] Step 31, model training phase;

[0022] Step 32: nonlinear event identification stage at the work site.

[0023] Furthermore, the model training phase includes the following steps:

[0024] Step 311: the model training phase begins;

[0025] Step 312: The high-definition surveillance camera collects video data;

[0026] Step 313: image pixel operation;

[0027] Step 314: Morphological opening operation;

[0028] Step 315: Find contour edge detection;

[0029] Step 316: Identify the minimum enclosing rectangle;

[0030] Step 317: Generate a data file with nonlinear event type annotations;

[0031] Step 318: Generate a data file with nonlinear events marked;

[0032] Step 319: Convert to TensorFlow format;

[0033] Step 3110: Under the condition of configuring hyperparameters, input the model network for training;

[0034] Step 3111: After training is completed, a pre-trained model file is obtained;

[0035] Step 3112: Convert to protocol buffer file;

[0036] Step 3113: Called by TensorFlow reasoning classification algorithm;

[0037] Step 3114: Model training is completed.

[0038] Furthermore, in step 313, the method of reducing the background influence of the video image by image pixel operation is as follows:

[0039] The image of the logistics park operation site is converted into a grayscale image, which is expressed by a mathematical formula: Set the threshold t, and divide the grayscale image of the logistics park operation site into two categories: event target and background. Figure 1 Dimensional histogram:

[0040] h(i) (i=1,2,…,L) (1)

[0041] Where L is the gray level, usually 255; let:

[0042]

[0043] Then the cross entropy discriminant function can be defined as:

[0044]

[0045] Where i is the grayscale value, t is the threshold value during thresholding, u(1,t) and u(t,L+1) are the intra-class means, which represent the grayscale of the event target and background after segmentation respectively;

[0046] The best segmentation threshold is to take the threshold t* that minimizes D(t), that is:

[0047] t * =argminD(t) (4)

[0048] In the actual algorithm implementation process, by traversing and calculating D(t*) with a threshold t* between 0 and L, the t* with the minimum value of D(t*) is found as the threshold to distinguish between the event target and the background in the logistics park operation site image.

[0049] Furthermore, in step 314, the morphological opening operation is to first perform an erosion operation on the event target region and then perform an expansion operation; wherein the erosion operation is to shrink the edge area of ​​the event target region and separate it from other adjacent event target regions, disconnect the narrow necks between the event target regions, and eliminate the elongated protruding areas of each event target region. The expansion operation is performed after the erosion operation to compensate for the area of ​​the event target region so as to improve the accuracy of the subsequent training process, specifically including:

[0050] A erosion operation and a dilation operation are combined, and a 5x5 cross-shaped structure element s is selected, with the anchor point at the center of the structure element s(3,3), as shown in the following matrix form:

[0051]

[0052] Corrosion operation: Use the structure element s to slide on the image f of the logistics park operation site image event target obtained in step 313, and correspond the anchor point position s(3,3) of the structure element s to the pixel position of the image f. The gray value of the target pixel is set to the area where the structure element value is 1, corresponding to the minimum value of the pixel in the image area; it can be expressed as:

[0053] dst(x, y) = min (x′，y′)：s(x′，y′)≠0 f(x+x′,y+y′) (6)

[0054] Where dst(x, y) is the resulting event target image after corrosion;

[0055] Dilation operation: Use the structure element s to slide on the target image dst after the erosion, and set the anchor point position s(3,3) of the structure element s to the pixel position of the image dst. The gray value of the target pixel is set to the area where the structure element value is 1, corresponding to the maximum value of the pixel in the image area; it can be expressed as follows:

[0056] g(x,y)=max(x′,y′): s(x′,y′)≠0dst(x+x′,y+y′) (7)

[0057] The final g(x,y) is the result of the target area of ​​the logistics park operation site event after a morphological opening operation combining a corrosion operation and a dilation operation.

[0058] Furthermore, in step 315, the contour edge detection is used to outline the area of ​​the event target of the logistics park operation site. Generally, the edge operator template that can be selected is convolved with the logistics park operation site event target area image after the morphological opening operation in step 3144. The Roberts operator, Prewitt operator, Sobel operator, Laplace operator, Log operator, and Canny operator can be selected.

[0059] Among them, the Laplace operator is used, which is expressed in mathematical formula as follows:

[0060]

[0061] Among them, f(x, y) is the pixel function of the event target area image of the logistics park operation site, is the gradient at the (x, y) position, and the magnitude of the gradient is expressed mathematically as:

[0062]

[0063] The above algorithms can be implemented by calling the image processing library of Hikvision's AI open platform Super Brain.

[0064] Furthermore, in step 316, the interval rotation method is used to identify the minimum area circumscribed rectangle. The algorithm principle is that for the minimum area circumscribed rectangle of the convex polygon, there must be at least one side that is rotated with each side of the original convex polygon to coincide with the X-axis. The circumscribed rectangle corresponding to each side is obtained and the area is minimized, and the minimum area circumscribed rectangle of the convex polygon is obtained by inversion; it can be expressed as a mathematical formula:

[0065]

[0066] The above algorithms can be implemented by calling the image processing library of Hikvision's AI open platform Super Brain.

[0067] Furthermore, in step 32, the nonlinear event identification stage at the work site includes the following steps:

[0068] Step 321: The high-definition surveillance camera collects video data;

[0069] Step 322: Input the TensorFlow model;

[0070] Step 323: Outputting the dynamic tracking data of the operation area;

[0071] Step 324: outputting nonlinear events;

[0072] Step 325: Output nonlinear event type.

[0073] Further, in step S5, the nonlinear event electronic grid virtual isolation includes the following steps:

[0074] Step 51: The association algorithm starts;

[0075] Step 52: Read non-linear event operation elements;

[0076] Step 53: using k data of the operation element set as initial cluster centers; the number of clusters k is set according to the tightness of virtual isolation;

[0077] Step 54: The remaining data is calculated to the distance of the cluster center and assigned to the nearest center cluster;

[0078] Step 55: Update the cluster center mean;

[0079] Step 56: Determine whether the cluster center has changed; if it is, that is, there is a change, jump back to step 54 and recalculate;

[0080] Step 57: Otherwise, there is no change, and the cluster center is output as the associated operation element;

[0081] Step 58: End, obtain other operations associated with the nonlinear event, and finally obtain a virtually isolated operation area.

[0082] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0083] The electronic grid system and method for nonlinear events at the operation site of a logistics park provided by the present invention have the following beneficial effects: using artificial intelligence recognition technology to dynamically track and identify nonlinear events, discover the occurrence of such nonlinear events as early as possible, evaluate their types and parameters, determine the associated operation areas, and achieve cross-regional electronic grid virtual isolation to reduce the negative impact of nonlinear events on other operations at the entire logistics park operation site and improve operation efficiency. The technology disclosed in the present invention and its adaptability improvement can be applied to various logistics scenarios such as e-commerce logistics, supermarket distribution, tobacco logistics, etc., and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] Figure 1 A structural diagram of a nonlinear event electronic grid system for a logistics park operation site provided by the present invention;

[0085] Figure 2 A flow chart of a nonlinear event electronic grid method for a logistics park operation site provided by the present invention;

[0086] Figure 3 This is a schematic diagram of nonlinear events at the logistics park operation site;

[0087] Figure 4 Adjust the process diagram for non-linear events at the logistics park operation site;

[0088] Figure 5 for Figure 2 The nonlinear event intelligent identification flow chart in the method flow chart shown;

[0089] Figure 6 for Figure 1 Schematic diagram of virtual isolation of nonlinear event electronic grid system in the method flow chart shown;

[0090] Figure 7 for Figure 2 Virtual isolation flow chart of the nonlinear event electronic grid method in the method flow chart shown. DETAILED DESCRIPTION

[0091] The specific implementation of the embodiment of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the embodiment of the present invention, and is not used to limit the embodiment of the present invention.

[0092] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0093] The present invention will be described in detail below in conjunction with exemplary embodiments with reference to the accompanying drawings.

[0094] Example 1

[0095] like Figure 2 As shown, the present invention provides a nonlinear event electronic grid method for a logistics park operation site, comprising the following steps:

[0096] Step S1: A high-definition surveillance camera collects video data 201 of the operation area;

[0097] Step S2: The video data is transmitted to the artificial intelligence dedicated server 202;

[0098] Step S3: The artificial intelligence dedicated server identifies nonlinear events 203;

[0099] Step S4: non-linear time-correlated region identification 204;

[0100] Step S5: Non-linear event electronic grid virtual isolation 205.

[0101] like Figure 3As shown, a schematic diagram of nonlinear events at the operation site of a logistics park is shown. A high-definition surveillance camera 300 performs real-time monitoring of the operation scene of the operation area A301, including intelligent identification and tracking of personnel, equipment, vehicles, and goods in the area. When a suspected nonlinear event 303 is identified, an electronic grid is set up for virtually isolating the associated operation area 304 affected by the nonlinear event 303. The associated operation area 304 may be in the same operation area A301 as the nonlinear event 303, or may be in an adjacent operation area B302 across regions. The nonlinear event 303 includes at least the following abnormal operation conditions: abnormal operator, failure of automatic equipment, abnormal operation of goods, etc. These abnormalities will cause the original operation schedule to be interrupted and increase the possibility of secondary abnormalities in the operation of the associated area. Therefore, the operation abnormality is intelligently identified, the operation of the associated area is isolated, and the operation process of other areas is adjusted to reduce the negative impact of the nonlinear event on the entire operation of the logistics park.

[0102] like Figure 4 As shown, a schematic diagram of the nonlinear event adjustment process of the operation site of the logistics park is shown, wherein the normal operation process includes: operation scheduling 401, operation 1 402 of the operation sequence of the operation area A, operation 2 403 of the operation demand of the operation area A, ..., operation n 404 of the operation sequence of the operation area A. The operation schedule 401 is optimized according to the status of the operator, equipment, vehicle, and goods, and the operation sequence is obtained, and the operation sequence is executed sequentially according to the pre-arranged operation schedule 401.

[0103] However, during the execution of the operation, nonlinear events often occur, which may cause the normal operation process to be interrupted, blocked, and other problems, and may also affect the status of related operators, equipment, vehicles, and goods, thereby causing the normal operation process in other areas to be interrupted, blocked, and other problems. Therefore, the nonlinear event adjustment process includes: operation scheduling 404, operation sequence 405 of operation area A, detection of nonlinear events in operation area A 406, isolation of operation area A 407, isolation of related operations 408, resolution of nonlinear events in operation area A 409, readjustment of operation scheduling 410, new operation sequence 411 of operation area A, and related new operation sequence 412. After the job scheduling 404 in the normal operation process, the operation sequence 405 of the operation area A executes each operation normally. After the non-linear event 406 of the operation area A is detected, the process is adjusted, including isolating the operation area A 407, and isolating the associated operations 408 in its associated areas. After the non-linear event 409 of the operation area A is resolved, the operation schedule 410 is readjusted according to the new status of the operators, equipment, vehicles, and goods, thereby forming a new operation sequence 411 of the operation area A and an associated new operation sequence 412.

[0104] like Figure 5 As shown, Figure 5 The nonlinear event intelligent identification flow chart of the nonlinear event electronic grid method for the operation site of the logistics park provided by the present invention is to mark the historical nonlinear events and event types of the monitoring video data, and train the artificial intelligence model to realize the intelligent identification of nonlinear events. The specific steps include two stages, namely: step 31 model training stage and step 32 operation site nonlinear event identification stage. The model training stage includes the following steps:

[0105] Step 3111: Model training phase starts 500;

[0106] Step 312: The high-definition surveillance camera collects video data 501;

[0107] Step 313: Image pixel operation 502;

[0108] Step 314: Morphological opening operation 503;

[0109] Step 315: Find contour edge detection 504;

[0110] Step 316: Identify the minimum circumscribed rectangle 505;

[0111] Step 317: Generate a data file 506 for nonlinear event type annotation;

[0112] Step 318: Generate a data file 507 after nonlinear event annotation;

[0113] Step 319: Perform TensorFlow format conversion 508;

[0114] Step 3110: Under the condition of configuring the hyperparameters 513, input the model network for training 509;

[0115] Step 3111: After training is completed, a pre-trained model file 510 is obtained;

[0116] Step 3112: Convert to protocol buffer file 511;

[0117] Step 3113: Called by TensorFlow reasoning classification algorithm 512;

[0118] Step 3114: Model training ends 514.

[0119] Specifically, model training uses the TensorFlow library, image, and video preprocessing library that comes with Hikvision's AI open platform Super Brain's artificial intelligence dedicated server.

[0120] Specifically, the model training phase is repeated 1000 times, and the recognition effect of the model is verified. The model obtained after training will be used in the nonlinear event recognition phase at the work site.

[0121] The image pixel operation 502 of step 313 reduces the background influence of the video image; generally speaking, when the pixel value difference between the background of the logistics park operation site and the event target is large, the pixel value difference between the event target and the background can be used to remove the background of the logistics park operation site. A simple method is to select a pixel value threshold, and pixels greater than the threshold are foreground, and pixels less than the threshold are background. However, in most cases, the ambient light intensity of the logistics park operation site is complex, and it is necessary to use an image segmentation algorithm to select an adaptive segmentation threshold. An embodiment of the present invention uses a minimum cross entropy algorithm to obtain the segmentation threshold. Cross entropy can be regarded as minimizing the cross entropy index between the segmented event target pixels and background pixels, that is, the expected value of the entire image information amount when the background pixels are replaced by event target pixels is minimized. Before processing, the logistics park operation site image is converted into a grayscale image, which is expressed by a mathematical formula: Let the threshold t divide the grayscale image of the logistics park operation site image into two categories: event target and background, and the grayscale Figure 1 The dimension histogram is:

[0122] h(i) (i=1,2,…,L) (1)

[0123] Where L is the gray level, usually 255; let:

[0124]

[0125] Then the cross entropy discriminant function can be defined as:

[0126]

[0127] Where i is the gray value, t is the threshold value during thresholding, u(1, t) and u(t, L+1) are the intra-class means, which represent the grayscale of the event target and background after segmentation, respectively. The best segmentation threshold is to take the threshold t* that minimizes D(t), that is:

[0128] t * =argmin D(t) (4)

[0129] In the actual algorithm implementation process, by traversing and calculating D(t*) with a threshold value t* between 0 and L, t* with the minimum value of D(t*) is found as the threshold for distinguishing event targets and backgrounds in the images of the logistics park operation site. This algorithm can be implemented by calling the image processing library of Hikvision AI open platform Ultra Brain.

[0130] The morphological opening operation 503 of step 314 is a method of image morphological analysis, and its purpose is to further distinguish the event targets in the logistics park operation site. The morphological opening operation is to first perform an erosion operation on the event target area, and then perform an expansion operation. The erosion operation is to shrink the edge area of ​​the event target area and separate it from other adjacent event target areas, disconnect the narrow necks between the event target areas, and eliminate the slender protruding areas of each event target area. After the erosion operation, the expansion operation is performed to compensate for the area of ​​the event target area, so as to improve the accuracy of the subsequent training process.

[0131] One embodiment of the present invention adopts the following morphological opening algorithm, including a combination of an erosion operation and a dilation operation, and selects a 5x5 cross-shaped structure element s, with the anchor point at the center of the structure element s(3,3), such as the following matrix form:

[0132]

[0133] Corrosion operation: Use the structure element s to slide on the image f of the logistics park operation site image event target obtained in step 3, and set the anchor point position s(3,3) of the structure element s to the pixel position of the image f. The gray value of the target pixel is set to the area where the structure element value is 1, corresponding to the minimum value of the pixel in the image area. It can be expressed as:

[0134] dst(x, y) = min (x′，y′)：s(x′ , y′)≠0 f(x+x′,y+y′) (6)

[0135] Where dst(x, y) is the resulting event target image after erosion.

[0136] Dilation operation: Use the structure element s to slide on the target image dst after the erosion, and set the anchor point position s(3,3) of the structure element s to the pixel position of the image dst. The gray value of the target pixel is set to the area where the structure element value is 1, corresponding to the maximum value of the pixel in the image area. It can be expressed as:

[0137] g(x,y)=max(x′,y′) ) : s(x′,y′)≠0dst(x+x′,y+y′) (7)

[0138] Finally, g(x,y) is the result of the target area of ​​the logistics park operation site event after a morphological opening operation combining a corrosion operation and a dilation operation. This algorithm can be implemented by calling the image processing library of Hikvision AI open platform Ultrain.

[0139] Among them, the search contour edge detection 504 of step 315 is used to outline the area of ​​the event target of the logistics park operation site. Generally, the edge operator template that can be selected is convolved with the logistics park operation site event target area image processed by the morphological opening operation 503 in step 4. The Roberts operator, Prewitt operator, Sobel operator, Laplace operator, Log operator, and Canny operator can be selected, each of which has its own advantages and disadvantages. One embodiment of the present invention selects the Laplace operator, which is expressed as a mathematical formula:

[0140]

[0141] Where f(x, y) is the pixel function of the event target area image at the logistics park operation site, is the gradient at the (x, y) position, and the magnitude of the gradient is expressed mathematically as:

[0142]

[0143] Similarly, the algorithm can be implemented by calling the image processing library of Hikvision's AI open platform Super Brain.

[0144] Specifically, in step 316, the identified minimum area bounding rectangle 505, the event areas at the logistics park operation site mostly overlap. In order to distinguish the respective areas more efficiently, after obtaining the contour, it is necessary to identify and extract the bounding rectangle with the smallest area. One embodiment of the present invention adopts the interval rotation method, and its algorithm principle is that for the minimum area bounding rectangle of the convex polygon, there must be at least one side that is rotated with each side of the original convex polygon to coincide with the X-axis. The bounding rectangle corresponding to each side is obtained and the area is the smallest, and then the minimum area bounding rectangle of the convex polygon is obtained by inversion. It can be expressed as a mathematical formula:

[0145]

[0146] This algorithm can be implemented by calling the image processing library of Hikvision's AI open platform Super Brain.

[0147] The nonlinear event identification stage at the work site includes the following steps:

[0148] Step 321: The high-definition surveillance camera collects video data 515;

[0149] Step 322: Input TensorFlow model 516;

[0150] Step 323: Outputting the operation area dynamic tracking data 517;

[0151] Step 324: outputting nonlinear events 518;

[0152] Step 325: Output nonlinear event type 519.

[0153] like Figure 6 As shown, Figure 6 A virtual isolation schematic diagram of an electronic grid system for nonlinear events at an operation site in a logistics park provided by the present invention is described in the figure. A high-definition monitoring camera 600 collects and intelligently identifies nonlinear events 603 in an operation area A 601, and detects that there is an associated operation in an operation area B 602, so an associated operation area 604 is set. After the operation scheduling server 605 obtains the information of the associated operation, the associated operation area 604 is virtually isolated, that is, according to the position information of the operation area A 601 and the operation area B 602, an electronic grid 606 for the operation area A and an electronic grid 607 for the operation area B are generated respectively, and a virtual isolation area 608 is marked accordingly, thereby correspondingly generating an operation element A 609 in the virtual isolation area and an operation element B 610 in the virtual isolation area, and correspondingly generating an operation element A 611 and an operation element B 612 outside the virtual isolation area that are not associated and are outside the virtual isolation area 608.

[0154] The virtual isolation area 608 is not a physical isolation measure, but a mark of a non-linear event in the scheduling system of the job scheduling server 605. The relevant job scheduling will be restarted after the non-linear event is resolved.

[0155] The operating element A 609 within the virtual isolation area, the operating element B 610 within the virtual isolation area, the operating element A 611 outside the virtual isolation area, and the operating element B 612 outside the virtual isolation area include operating elements such as operators, equipment, vehicles, goods, locations, time, etc., or a combination thereof.

[0156] like Figure 7 As shown, Figure 7 The invention provides a virtual isolation flow chart of an electronic grid method for nonlinear events at a logistics park operation site, comprising the following steps:

[0157] Step 51: The association algorithm starts 701;

[0158] Step 52: Read the non-linear event operation element 702;

[0159] Step 53: taking k data of the operation element set as the initial cluster center 703;

[0160] Step 54: The remaining data is calculated to have a distance to the cluster center and is assigned to the nearest center cluster 704;

[0161] Step 55: Update the cluster center mean 705;

[0162] Step 56: Determine whether the cluster center has changed 706? If it is determined that there is a change, then jump back to step 54 and recalculate;

[0163] Step 57: Otherwise, there is no change, and the cluster center is output as the associated operation element 707;

[0164] Step 58: End 708.

[0165] The algorithm obtains other operations associated with nonlinear events and finally obtains a virtually isolated operation area.

[0166] The number of clusters k is set according to the tightness of virtual isolation.

[0167] exist Figure 6 There are four operations Oi(h, d, v, g, p, t) (i = 1, 2, 3, 4), each of which has its own operation elements: operator h, equipment d, vehicle v, cargo g, location p, time t. Example 1 Select parameter k = 2, that is, there are at most 2 associated operations in the virtual isolation area, select O1 and O2 as the initial clustering centers, and calculate the distance from O3 to O1 and O2 If the operator involved in the two operations is the same person, the difference is 0, otherwise the difference is 1; if the equipment involved in the two operations is the same equipment, the difference is 0, otherwise the difference is 1; if the vehicle involved in the two operations is the same vehicle, the difference is 0, otherwise the difference is 1; if the goods involved in the two operations are the same goods, the difference is 0, otherwise the difference is 1; the difference between the two operation locations is the spatial distance between the two operation areas; if the two operations overlap in time, the difference is 0, otherwise the difference is 1. O3 is close to O1, so it is closer to O1, and the cluster center is adjusted to between O3 and O1. Repeat the calculation of O4, and the output cluster center is the associated operation, and the nonlinear events that fall into the cluster are classified into the virtual isolation operation area.

[0168] Example 2

[0169] like Figure 1As shown, the present invention provides a nonlinear event electronic grid system for a logistics park operation site, comprising: a plurality of high-definition monitoring cameras 101, 102, 103, 104, 105, 106, which divide the logistics park operation site 100 into a plurality of operation areas 107, 108, 109, 110, 111, 112, each of the high-definition monitoring cameras 101, 102, 103, 104, 105, 106 collects video data of the operation scenes of the corresponding operation areas 107, 108, 109, 110, 111, 112, respectively, and is connected to an artificial intelligence dedicated server 113, and the video data is uploaded to the artificial intelligence dedicated server 113 for nonlinear event intelligent recognition, and the artificial intelligence dedicated server 113 is connected to a job scheduling server 114, and the nonlinear event intelligent recognition result is sent to the job scheduling server 114 for nonlinear event electronic grid virtual isolation, thereby guiding the optimization and adjustment of the nonlinear event process;

[0170] Preferably, the high-definition surveillance cameras 101, 102, 103, 104, 105, and 106 are Hikvision 2 million 1 / 2.7" CMOS black light-level cylindrical network cameras with built-in hybrid fill light, which can adjust the power of infrared light and white light, and are suitable for the all-weather tracking and identification needs of the logistics park;

[0171] Preferably, the artificial intelligence dedicated server 113 uses Hikvision AI open platform super brain, which has 4 built-in GUP chips, supports detection algorithms, classification algorithms, and supports SDK development of mainstream artificial intelligence network models such as TensorFlow.

[0172] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. An electronic grid method for nonlinear events at a logistics park operation site, characterized in that: The method comprises the following steps: Step S1: A high-definition surveillance camera collects video data of the operation area; Step S2: The video data is transmitted to a dedicated artificial intelligence server; Step S3: The artificial intelligence dedicated server identifies nonlinear events; Step S4: nonlinear time-related region identification; Step S5: Virtual isolation of nonlinear event electronic grid.

2. The electronic grid method for nonlinear events at a logistics park operation site as claimed in claim 1, characterized in that: In step S3, when a suspected nonlinear event is identified, an electronic grid is set up for virtually isolating the associated operation area affected by the nonlinear event. The associated operation area may be in the same operation area A as the nonlinear event, or may be across regions in an adjacent operation area B. The nonlinear event includes at least the following abnormal operation conditions: operator abnormality, automatic equipment failure, and operation cargo abnormality.

3. The electronic grid method for nonlinear events at a logistics park operation site as claimed in claim 1, characterized in that: In step S3, the historical nonlinear events and event types of the surveillance video data are annotated, and an artificial intelligence model is trained to realize intelligent recognition of nonlinear events. The specific steps include two stages, namely: Step 31, model training phase; Step 32: nonlinear event identification stage at the work site.

4. The electronic grid method for nonlinear events at a logistics park operation site as claimed in claim 3, characterized in that: The model training phase includes the following steps: Step 311: the model training phase begins; Step 312: The high-definition surveillance camera collects video data; Step 313: image pixel operation; Step 314: Morphological opening operation; Step 315: Find contour edge detection; Step 316: Identify the minimum enclosing rectangle; Step 317: Generate a data file with nonlinear event type annotations; Step 318: Generate a data file with nonlinear events marked; Step 319: Convert to TensorFlow format; Step 3110: Under the condition of configuring hyperparameters, input the model network for training; Step 3111: After training is completed, a pre-trained model file is obtained; Step 3112: Convert to protocol buffer file; Step 3113: Called by TensorFlow reasoning classification algorithm; Step 3114: Model training is completed.

5. The electronic grid method for nonlinear events at a logistics park operation site as claimed in claim 4, characterized in that: The method for reducing the background effect of the video image by image pixel operation in step 313 is as follows: The image of the logistics park operation site is converted into a grayscale image, which is expressed by a mathematical formula: Set the threshold t, divide the grayscale image of the logistics park operation site into two categories: event target and background, and the one-dimensional histogram of the grayscale image: h(i)(i=1,2,…,L) (1) Where L is the gray level, usually 255; let: Then the cross entropy discriminant function can be defined as: Where i is the grayscale value, t is the threshold value during thresholding, u(1,t) and u(t,L+1) are the intra-class means, which represent the grayscale of the event target and background after segmentation respectively; The best segmentation threshold is to take the threshold t* that minimizes D(t), that is: t * =arg min D(t) (4) In the actual algorithm implementation process, by traversing and calculating D(t*) with a threshold t* between 0 and L, the t* with the minimum value of D(t*) is found as the threshold to distinguish between the event target and the background in the logistics park operation site image.

6. The electronic grid method for nonlinear events at a logistics park operation site as claimed in claim 4, characterized in that: In step 314, the morphological opening operation is to first perform an erosion operation on the event target region and then perform an expansion operation; wherein the erosion operation is to shrink the edge area of ​​the event target region and separate it from other adjacent event target regions, disconnect the narrow necks between the event target regions, and eliminate the elongated protruding areas of each event target region. After the erosion operation, the expansion operation is performed to compensate for the area of ​​the event target region in order to improve the accuracy of the subsequent training process, which specifically includes: A erosion operation and a dilation operation are combined, and a 5x5 cross-shaped structure element s is selected, and the anchor point is at the center of the structure element s(3,3), such as the following matrix form: Corrosion operation: Use the structure element s to slide on the image f of the logistics park operation site image event target obtained in step 313, and correspond the anchor point position s(3,3) of the structure element s to the pixel position of the image f. The gray value of the target pixel is set to the area where the structure element value is 1, corresponding to the minimum value of the pixel in the image area; it can be expressed as: dst(x,y)=min (x′,y′):s(x′,y′)≠0 f(x+x′,y+y′) (6) Where dst(x,y) is the resulting event target image after corrosion; Dilation operation: Use the structure element s to slide on the target image dst after the erosion, and set the anchor point position s(3,3) of the structure element s to the pixel position of the image dst. The gray value of the target pixel is set to the area where the structure element value is 1, corresponding to the maximum value of the pixel in the image area; it can be expressed as: g(x,y) = max (x′,y′):s(x′,y′)≠0 dst(x+x′,y+y′) (7) The final g(x,y) is the result of the target area of ​​the logistics park operation site event after a morphological opening operation combining a corrosion operation and a dilation operation.

7. The electronic grid method for nonlinear events at a logistics park operation site as claimed in claim 4, characterized in that: In step 315, the contour edge detection is used to outline the area of ​​the event target of the logistics park operation site. Generally, the edge operator template that can be selected is convolved with the logistics park operation site event target area image after the morphological opening operation in step 3144. The Roberts operator, Prewitt operator, Sobel operator, Laplace operator, Log operator, and Canny operator can be selected; Among them, the Laplace operator is used, which is expressed in mathematical formula as follows: Among them, f(x,y) is the pixel function of the event target area image of the logistics park operation site, is the gradient at the (x,y) position, and the magnitude of the gradient is expressed mathematically as: The above algorithms can be implemented by calling the image processing library of Hikvision's AI open platform Super Brain.

8. The electronic grid method for nonlinear events at a logistics park operation site as claimed in claim 4, characterized in that: In step 316, the minimum area circumscribed rectangle is identified by using the interval rotation method. The algorithm principle is that for the minimum area circumscribed rectangle of the convex polygon, there must be at least one side that is rotated with each side of the original convex polygon to coincide with the X-axis. The circumscribed rectangle corresponding to each side is obtained and the area is minimized. The minimum area circumscribed rectangle of the convex polygon is obtained by inverting. It can be expressed as follows: The above algorithms can be implemented by calling the image processing library of Hikvision's AI open platform Super Brain.

9. The electronic grid method for nonlinear events at a logistics park operation site as claimed in claim 3, characterized in that: In step 32, the nonlinear event identification stage at the work site includes the following steps: Step 321: The high-definition surveillance camera collects video data; Step 322: Input the TensorFlow model; Step 323: Outputting the dynamic tracking data of the operation area; Step 324: outputting nonlinear events; Step 325: Output nonlinear event type.

10. The electronic grid method for nonlinear events at a logistics park operation site as claimed in claim 1, characterized in that: In step S5, the nonlinear event electronic grid virtual isolation includes the following steps: Step 51: The association algorithm starts; Step 52: Read non-linear event operation elements; Step 53: using k data of the operation element set as initial cluster centers; the number of clusters k is set according to the tightness of virtual isolation; Step 54: The remaining data is calculated to the distance of the cluster center and assigned to the nearest center cluster; Step 55: Update the cluster center mean; Step 56: Determine whether the cluster center has changed; if it is, that is, there is a change, jump back to step 54 and recalculate; Step 57: Otherwise, there is no change, and the cluster center is output as the associated operation element; Step 58: End, obtain other operations associated with the nonlinear event, and finally obtain a virtually isolated operation area.

Images