Warehouse safety implementation method and system based on machine vision

Abstract

The invention discloses a warehouse safety implementation method and system based on machine vision, belongs to the technical field of warehouse supervision, and is used for solving the technical problems that the existing warehouse safety supervision method is easy to cause monitoring dead angles, new monitoring dead angles are possibly generated in the cargo flowing process, and potential safety hazards are brought to warehouse supervision. The method comprises the following steps: acquiring a monitoring picture image of a plurality of cameras arranged in a warehouse, and performing boundary detection on a ground part in the monitoring picture image to obtain a ground area; performing cross detection on a ground area in each monitoring picture image to generate a non-blind area placement area map; and when new goods are warehoused, placing the goods according to the non-blind area placing area map, carrying out monitoring blind area detection on the placed goods, giving an alarm if a new monitoring blind area is detected, and recommending a position for additionally arranging a camera so as to eliminate the new monitoring blind area. Visual hindrance of improper operation and potential safety hazards in warehouse detection is reduced, and the warehouse supervision effect is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of warehouse supervision, and particularly to a method and system for realizing warehouse safety based on machine vision. Background Art

[0002] The warehouse environment is complex, involving a large number of goods and equipment, so potential safety hazards often exist. For example, factors such as non-standard operations, old equipment or lack of regular maintenance may all lead to the occurrence of warehouse safety accidents, threatening the lives of employees. Moreover, the goods stored in the warehouse often have high value. If theft and other behaviors occur, it will also bring significant economic losses and legal risks to the enterprise.

[0003] To avoid the occurrence of the above situations, warehouses often monitor various situations in the warehouse through surveillance cameras or manual supervision. However, even when supervised by surveillance cameras, it still requires manual supervision in front of the monitor, which not only wastes labor costs and time costs, but also is prone to missed viewing, which is not conducive to the safety of the warehouse. In addition, the environment in the warehouse is relatively complex, and it is easy to have monitoring dead corners and blind spots. The situations occurring in these dead corners cannot be seen during monitoring, bringing potential safety hazards to warehouse supervision. In addition, even if the monitoring dead corners are initially eliminated by reasonably arranging the positions of surveillance cameras, due to the flow of goods in the warehouse, there are processes such as outbound and inbound, and the placement positions of the goods entering the warehouse each time are not necessarily fixed. Therefore, new monitoring dead corners may be generated during the flow of goods in the warehouse, affecting the safety of the warehouse. Summary of the Invention

[0004] Embodiments of the present invention provide a method and system for realizing warehouse safety based on machine vision to solve the following technical problems: The existing warehouse safety supervision methods are prone to monitoring dead corners, and new monitoring dead corners may be generated during the flow of goods, bringing potential safety hazards to warehouse supervision.

[0005] Embodiments of the present invention adopt the following technical solutions:

[0006] On the one hand, embodiments of the present invention provide a method for realizing warehouse safety based on machine vision. The method includes: acquiring the monitoring screen images of a number of cameras arranged in the warehouse, and performing boundary detection on the ground part in the monitoring screen images to obtain the ground area; performing cross detection on the ground areas in each monitoring screen image to generate a non-blind area placement map.

[0007] When new goods are put into storage, place the goods according to the non-blind area placement map, and perform monitoring blind area detection on the placed goods. If a new monitoring blind area is detected, an alarm is issued, and the position for adding a camera is recommended to eliminate the new monitoring blind area.

[0008] In a feasible implementation, monitor screen images of multiple cameras deployed in the warehouse are obtained, and boundary detection is performed on the ground part in the monitor screen images to obtain the ground area, specifically including:

[0009] Obtain monitor screen images of multiple cameras that have been deployed in the current warehouse;

[0010] Perform color cast detection and correction on the monitor screen images to obtain standard monitor images;

[0011] Perform foreground object recognition on the corrected monitor screen images, and segment the foreground objects from the background part to obtain foreground object images and background area images;

[0012] Divide the monitor screen images detecting the same foreground object into a group of associated image groups, and divide the corresponding cameras into a group of associated camera groups;

[0013] Perform boundary detection on the background area images in an associated image group, identify the ground area, and obtain the position coordinates of the boundary points of the ground area in the world coordinate system.

[0014] In a feasible implementation, performing color cast detection and correction on the monitor screen images to obtain standard monitor images specifically includes:

[0015] Preprocess the monitor screen images through Gaussian filtering, and convert the preprocessed images from the RGB color space to the YUV color space;

[0016] Calculate the pixel average variance of the UV channels in the monitor screen images in the YUVA color space to obtain the color cast factor;

[0017] If the color cast factor is less than the first preset threshold, determine that the monitor screen image is a color cast image;

[0018] Convert the color cast image from the RGB channels to the hue channel H of the HSL color space to obtain the hue distribution map of the color cast image;

[0019] Based on the hue distribution map, determine the color deviation information of the color cast image; wherein, the color deviation information at least includes the deviated color and the degree of color cast;

[0020] According to the color deviation information, perform image correction on the color cast image to obtain the standard monitor image.

[0021] In a feasible implementation, according to the color deviation information, performing image correction on the color cast image to obtain the standard monitor image specifically includes:

[0022] Determine the confidence of each row of pixels in the offset-color image in the UV channel based on the size of the correction window;

[0023] Determine the target value of the correction result according to the degree of color offset of the offset-color image;

[0024] Determine the correction coefficient of each row of pixels according to the confidence of each row of pixels in the UV channel of the offset-color image, the pixel average value, the size of the correction window, and the target value of the correction result;

[0025] Perform line-by-line correction on each row of pixels through the correction coefficient of each row of pixels, and restore it to an RGB channel image to obtain the standard monitoring image.

[0026] In a feasible implementation manner, perform foreground target recognition on the corrected monitoring screen image, and segment the foreground target from the background part to obtain a foreground target image and a background area image, specifically including:

[0027] Collect close-up images of various foreground targets that appear in the warehouse scene, and construct a foreground target training data set;

[0028] Build a foreground target recognition model based on the PP-YOLO network, and construct several detection candidate boxes with different sizes in the foreground target recognition model;

[0029] Input the foreground target training data set into the foreground target recognition model, perform foreground target detection through the several detection candidate boxes with different sizes, and train the foreground target recognition model;

[0030] Perform foreground target recognition on the corrected monitoring screen image through the trained foreground target recognition model to obtain the current foreground target;

[0031] In the corrected monitoring screen image, segment all the current foreground targets from the rest to obtain a foreground target image and a background area image.

[0032] In a feasible implementation manner, perform boundary detection on the background area image in an associated image group, identify the ground area, and obtain the position coordinates of the boundary points of the ground area in the world coordinate system, specifically including:

[0033] Obtain the edge feature map corresponding to each background area image in the associated image group through edge detection technology;

[0034] Based on a pre-created ground feature extraction module, extract ground edge features from the edge feature map to identify the corresponding ground area;

[0035] Obtain the boundary line of the ground area in each background area image, and uniformly select the two-dimensional coordinates of several boundary points in the boundary line;

[0036] Project the two-dimensional coordinates into the world coordinate system, obtain the position coordinates of the boundary points in the world coordinate system, and obtain the ground boundary position coordinate set of each background area image.

[0037] In a feasible implementation manner, perform cross-detection on the ground areas in each monitoring screen image to generate a non-blind area placement area map, which specifically includes:

[0038] Sort the background area images according to the order of the monitoring fields of view of the cameras corresponding to each background area image in the associated image group from large to small;

[0039] Select the first background area image and the second background area image, and calculate the intersection area of the ground areas of the two according to the corresponding ground boundary position coordinate sets;

[0040] Calculate the intersection area of the third background area image and the intersection area, and so on, until the area of the intersection area is lower than a preset area threshold, and stop the iteration;

[0041] Obtain the intersection area at this time, determine it as the non-blind area placement area, and generate the non-blind area placement area map in the warehouse top view according to the boundary position coordinates.

[0042] In a feasible implementation manner, perform monitoring blind area detection on the placed goods, and issue an alarm if a new monitoring blind area is detected, which specifically includes:

[0043] Obtain the associated camera group corresponding to the non-blind area placement area map of the new goods placement;

[0044] Obtain the real-time monitoring image of the new goods through the associated camera group, and extract the target image area of the new goods in the real-time monitoring image;

[0045] Stitch the target image areas captured by each camera in the associated camera group to obtain the appearance stitched image of the new goods;

[0046] Compare the appearance stitched image with the appearance scanned image after the new goods are put into storage. If the coincidence rate of the two is lower than the second preset threshold, it is determined that there is a new monitoring blind area for the new goods, and an alarm is sent to the management terminal.

[0047] In a feasible implementation manner, recommend the positions for adding cameras to eliminate the new monitoring blind area, which specifically includes:

[0048] Determine the position of the new goods that have not been captured by the camera according to the comparison result between the appearance splicing image and the appearance scanning image;

[0049] Determine the recommended position for adding a camera according to the orientation of this position to eliminate the new monitoring blind area.

[0050] On the other hand, an embodiment of the present invention also provides a warehouse security implementation system based on machine vision, and the system includes:

[0051] A goods placement position recommendation module, configured to obtain the monitoring screen images of multiple cameras arranged in the warehouse, perform boundary detection on the ground part in the monitoring screen images to obtain the ground area; perform cross-detection on the ground areas in each monitoring screen image to generate a non-blind area placement area map;

[0052] A goods placement blind area detection module, configured to, when new goods are put into storage, place the goods according to the non-blind area placement area map, and perform monitoring blind area detection on the placed goods. If a new monitoring blind area is detected, an alarm is issued, and the recommended position for adding a camera is provided to eliminate the new monitoring blind area.

[0053] Compared with the prior art, a warehouse security implementation method and system based on machine vision provided by an embodiment of the present invention have the following beneficial effects:

[0054] The present invention generates a placement position area map with no blind area or a small blind area for the goods that have not been put into storage in the warehouse through machine vision technology. Staff can place the goods according to the generated area map, so as to avoid new monitoring dead corners as much as possible without changing the current camera position. However, due to the height of the goods, in some cases, part of the space will still be blocked. Therefore, the present invention will identify whether there is a blind area again after the goods are actually put into storage. If the blind area cannot be eliminated by adjusting the angle of the camera, the best direction for adding a camera is calculated through the provided algorithm and recommended to the warehouse management staff. Through the above solution, it is possible to ensure that there is no monitoring blind area in the warehouse when each batch of goods is put into storage, thereby reducing the visual obstacles for detecting improper operations and potential safety hazards in the warehouse and improving the effect of warehouse supervision. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts. In the drawings:

[0056] Figure 1A flowchart of a method for realizing warehouse security based on machine vision provided by an embodiment of the present invention;

[0057] Figure 2 A schematic structural diagram of a system for realizing warehouse security based on machine vision provided by an embodiment of the present invention. Detailed implementation manners

[0058] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of this specification, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0059] An embodiment of the present invention provides a method for realizing warehouse security based on machine vision. As Figure 1 shown, the method for realizing warehouse security based on machine vision specifically includes steps S101 - S102:

[0060] S101. Obtain the monitoring screen images of several cameras arranged in the warehouse, and perform boundary detection on the ground part in the monitoring screen images to obtain the ground area.

[0061] Specifically, first obtain the monitoring screen images of multiple cameras already arranged in the current warehouse. Since the cameras in the warehouse are exposed to the air environment for a long time, they are prone to accidents such as pollution and collision, or due to reasons such as the aging of electronic components caused by long - term continuous operation, the probability of color deviation faults increases greatly. And cameras with color deviation faults cannot reflect the real colors in the scene, which will have a negative impact on subsequent target recognition and other processes. Therefore, the present invention first performs color deviation detection and correction on the monitoring screen images to obtain standard monitoring images. The specific implementation steps are as follows:

[0062] First, pre - process the monitoring screen images through Gaussian filtering technology, and convert the pre - processed images from the RGB color space to the YUV color space. Then calculate the pixel average variance of the UV channels in the YUVA color space of the monitoring screen images as the color deviation factor. If the color deviation factor is less than the first preset threshold, it is determined that the monitoring screen image is a color - deviated image. Then convert the color - deviated image from the RGB channels to the hue channel H of the HSL color space to obtain the hue distribution map of the color - deviated image. Thus, based on the hue distribution map, determine the color deviation information of the color - deviated image, where the color deviation information at least includes the deviated color and the degree of color deviation.

[0063] Further, according to the above color bias information, perform image correction on the color-biased image to obtain a standard monitoring image. The specific steps are as follows:

[0064] (1) Based on the size of the correction window, determine the confidence of each row of pixels in the color-biased image in the UV channels; this confidence can be a uniform coefficient x i , which is obtained through the following formula: where w is the size of the correction window.

[0065] (2) According to the color bias degree of the color-biased image, determine the target value A of the correction result. The target value of the correction result is a manually set value, which can be set according to the desired correction effect. Then, according to the confidence of each row of pixels in the UV channels of the color-biased image, the pixel average value, the size of the correction window, and the target value of the correction result, determine the correction coefficient of each row of pixels. The correction coefficient is obtained through the following formula: where is the pixel average value of the i-th row of pixels in the UV channels.

[0066] (3) Multiply the correction coefficient of each row of pixels by the pixel value of the corresponding row, thereby performing row-by-row correction on the color-biased image, and restore the corrected image to an RGB channel image to obtain a standard monitoring image.

[0067] Further, perform foreground object recognition on the corrected monitoring screen image, and segment the foreground object from the background part to obtain a foreground object image and a background area image. The specific implementation steps are as follows:

[0068] Collect close-up images of various foreground objects that appear in the warehouse scene to construct a foreground object training dataset. Then, build a foreground object recognition model based on the PP-YOLO network, and construct several detection candidate boxes of different sizes in the foreground object recognition model. Then, input the foreground object training dataset into the foreground object recognition model, and perform foreground object detection through several detection candidate boxes of different sizes to train the foreground object recognition model. Through the trained foreground object recognition model, perform foreground object recognition on the corrected monitoring screen image to obtain the current foreground object. In the corrected monitoring screen image, segment all the current foreground objects from the rest to obtain a foreground object image and a background area image.

[0069] Further, divide the monitoring screen images that detect the same foreground object into a group of associated image groups, and divide the corresponding cameras into a group of associated camera groups.

[0070] For example, if the cargo target A is recognized in the surveillance images captured by cameras 1, 3, 7, and 8, then cameras 1, 3, 7, and 8 are grouped into an associated camera group, and the surveillance images they capture are also grouped into an associated image group for subsequent image analysis.

[0071] Further, perform boundary detection on the background region images in an associated image group respectively, identify the ground region, and obtain the position coordinates of the boundary points of the ground region in the world coordinate system, specifically including:

[0072] Through edge detection technology, obtain the edge feature map corresponding to each background region image in the associated image group. Then, based on the pre-created ground feature extraction module, extract the ground edge features in the edge feature map to identify the corresponding ground region. Obtain the boundary line of the ground region in each background region image, and uniformly select the two-dimensional coordinates of several boundary points in the boundary line. Project these two-dimensional coordinates into the world coordinate system to obtain the position coordinates of the boundary points in the world coordinate system, and obtain the set of ground boundary position coordinates of each background region image.

[0073] S102. Perform cross-detection on the ground regions in each surveillance image to generate a non-blind area placement region map.

[0074] Specifically, arrange the background region images according to the order of the monitoring fields of view of the cameras corresponding to each background region image in the associated image group from large to small.

[0075] Further, select the first background region image and the second background region image, and calculate the intersection region of their ground regions according to the corresponding set of ground boundary position coordinates. Then calculate the intersection region of the third background region image and the intersection region, and so on, until the area of the intersection region is lower than the preset area threshold, and stop the iteration.

[0076] Obtain the intersection region at this time, determine it as the non-blind area placement region, and generate a non-blind area placement region map in the top view of the warehouse according to the boundary position coordinates.

[0077] S103. When new goods are put into storage, place the goods according to the non-blind area placement region map, and perform monitoring blind area detection on the placed goods. If a new monitoring blind area is detected, issue an alarm and recommend the position of adding cameras to eliminate the new monitoring blind area.

[0078] Specifically, after the staff place the goods according to the instructions of the generated non-blind area placement region map, obtain the real-time monitoring images of the new goods through the associated camera group corresponding to the non-blind area placement region map, and extract the target image region of the new goods from the real-time monitoring images.

[0079] Stitch the target image areas captured by each camera in the associated camera group to obtain an appearance stitched image of the new goods.

[0080] Compare the appearance stitched image with the appearance scanned image after the new goods are stored in the warehouse. If the coincidence rate between the two is lower than the second preset threshold, it is determined that there is a new monitoring blind area for the new goods, and an alarm is sent to the management terminal.

[0081] As a feasible implementation manner, according to the comparison result between the appearance stitched image and the appearance scanned image, determine the positions on each surface of the new goods that are not captured by the camera. According to the orientation of this position, determine the recommended positions for adding cameras to eliminate the new monitoring blind area.

[0082] In addition, the embodiment of the present invention also provides a warehouse security implementation system based on machine vision, as Figure 2 shown. The warehouse security implementation system 200 based on machine vision specifically includes:

[0083] The goods placement position recommendation module 210 is used to obtain the monitoring screen images of multiple cameras arranged in the warehouse, perform boundary detection on the ground part in the monitoring screen images to obtain the ground area; perform cross-detection on the ground areas in each monitoring screen image to generate a non-blind area placement area map.

[0084] The goods placement blind area detection module 220 is used to place the goods according to the non-blind area placement area map when the new goods are stored in the warehouse, and perform monitoring blind area detection on the placed goods. If a new monitoring blind area is detected, an alarm is issued and the recommended positions for adding cameras are recommended to eliminate the new monitoring blind area.

[0085] Each embodiment in the present invention is described in a progressive manner. The same or similar parts between each embodiment can be referred to each other, and the key points of each embodiment are the differences from other embodiments. In particular, for the embodiments of the device, equipment, and non-volatile computer storage medium, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.

[0086] The above describes specific embodiments of the present invention. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0087] The above are only the embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, various modifications and variations can be made to the embodiments of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the embodiments of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for realizing warehouse security based on machine vision, characterized in that, The method includes: Obtaining the monitoring screen images of a number of cameras deployed in the warehouse, performing boundary detection on the ground part in the monitoring screen images to obtain the ground area; performing cross-detection on the ground areas in each monitoring screen image to generate a non-blind area placement map; When new goods are put into storage, placing the goods according to the non-blind area placement map, and performing monitoring blind area detection on the placed goods. If a new monitoring blind area is detected, an alarm is issued, and the position for adding a camera is recommended to eliminate the new monitoring blind area.

2. The method for realizing warehouse safety based on machine vision according to claim 1, wherein, Obtaining the monitoring screen images of multiple cameras deployed in the warehouse, and performing boundary detection on the ground part in the monitoring screen images to obtain the ground area, specifically including: Obtaining the monitoring screen images of multiple cameras already deployed in the current warehouse; Performing color cast detection and correction on the monitoring screen images to obtain standard monitoring images; Performing foreground object recognition on the corrected monitoring screen images, and segmenting the foreground objects from the background part to obtain foreground object images and background area images; Dividing the monitoring screen images in which the same foreground object is detected into a group of associated image groups, and dividing the corresponding cameras into a group of associated camera groups; Performing boundary detection on the background area images in an associated image group, identifying the ground area, and obtaining the position coordinates of the boundary points of the ground area in the world coordinate system.

3. The method for realizing warehouse safety based on machine vision according to claim 2, wherein, Performing color cast detection and correction on the monitoring screen images to obtain standard monitoring images, specifically including: Preprocessing the monitoring screen images through Gaussian filtering, and converting the preprocessed images from the RGB color space to the YUV color space; Calculating the pixel average variance of the UV channels in the monitoring screen images in the YUVA color space to obtain the color cast factor; If the color cast factor is less than the first preset threshold, determining that the monitoring screen image is a color cast image; Converting the color cast image from the RGB channels to the hue channel H of the HSL color space to obtain the hue distribution map of the color cast image; Based on the hue distribution map, determining the color deviation information of the color cast image; wherein, the color deviation information at least includes the deviated color and the degree of color cast; Performing image correction on the color cast image according to the color deviation information to obtain the standard monitoring image.

4. The method for realizing warehouse safety based on machine vision according to claim 3, characterized in that, Performing image correction on the color cast image according to the color deviation information to obtain the standard monitoring image, specifically including: Determining the confidence of each row of pixels in the UV channels of the color cast image based on the size of the correction window; Determining the correction result target value according to the degree of color cast of the color cast image; Determining the correction coefficient of each row of pixels according to the confidence and pixel average value of each row of pixels in the UV channels of the color cast image, the size of the correction window, and the correction result target value; Performing row-by-row correction on each row of pixels through the correction coefficient of each row of pixels and restoring it to an RGB channel image to obtain the standard monitoring image.

5. The method for realizing warehouse safety based on machine vision according to claim 2, wherein, Performing foreground object recognition on the corrected monitoring screen images, and segmenting the foreground objects from the background part to obtain foreground object images and background area images, specifically including: Collect close-up images of various foreground objects that appear in the warehouse scene to construct a foreground object training dataset; Construct a foreground object recognition model based on the PP-YOLO network, and construct several detection candidate boxes of different sizes in the foreground object recognition model; Input the foreground object training dataset into the foreground object recognition model, perform foreground object detection through the several detection candidate boxes of different sizes, and train the foreground object recognition model; Perform foreground object recognition on the corrected surveillance video image through the trained foreground object recognition model to obtain the current foreground object; In the corrected surveillance video image, segment all the current foreground objects from the rest to obtain a foreground object image and a background region image.

6. The method for realizing warehouse safety based on machine vision according to claim 2, wherein, Perform boundary detection on the background region images in an associated image group, identify the ground region, and obtain the position coordinates of the boundary points of the ground region in the world coordinate system. Specifically, it includes: Through edge detection technology, obtain the edge feature map corresponding to each background region image in the associated image group; Based on a pre-created ground feature extraction module, extract ground edge features from the edge feature map to identify the corresponding ground region; Obtain the boundary line of the ground region in each background region image, and uniformly select the two-dimensional coordinates of several boundary points from the boundary line; Project the two-dimensional coordinates into the world coordinate system to obtain the position coordinates of the boundary points in the world coordinate system, and obtain the ground boundary position coordinate set of each background region image.

7. The method for realizing warehouse safety based on machine vision according to claim 6, wherein, Perform cross detection on the ground regions in each surveillance video image to generate a non-blind area placement region map. Specifically, it includes: Sort the background region images according to the order of the monitoring fields of view of the cameras corresponding to each background region image in the associated image group from large to small; Select the first background region image and the second background region image, and calculate the intersection region of the two ground regions according to the corresponding ground boundary position coordinate sets; Calculate the intersection region of the third background region image and the intersection region, and so on, until the area of the intersection region is lower than a preset area threshold, and stop the iteration; Obtain the intersection region at this time, determine it as the non-blind area placement region, and generate the non-blind area placement region map in the warehouse top view according to the boundary position coordinates.

8. A method for realizing warehouse safety based on machine vision according to claim 1, characterized in that, Perform monitoring blind area detection on the placed goods, and issue an alarm if a new monitoring blind area is detected. Specifically, it includes: Obtain the associated camera group corresponding to the non-blind area placement region map of the new goods placement; Obtain the real-time monitoring image of the new goods through the associated camera group, and extract the target image region of the new goods from the real-time monitoring image; Stitch the target image regions captured by each camera in the associated camera group to obtain an appearance stitched image of the new goods; Compare the appearance stitched image with the appearance scanned image after the new goods are put into storage. If the coincidence rate of the two is lower than a second preset threshold, it is determined that there is a new monitoring blind area for the new goods, and an alarm is sent to the management terminal.

9. The method for realizing warehouse security based on machine vision according to claim 8, characterized in that, Recommend the position to add a camera to eliminate the new monitoring blind area. Specifically, it includes: Determine the position of the new goods that have not been captured by the camera according to the comparison result between the appearance splicing image and the appearance scanning image; Determine the recommended position for adding a camera according to the orientation of this position to eliminate the new monitoring blind area.

10. A warehouse security implementation system based on machine vision, characterized in that, The system includes: A goods placement position recommendation module, configured to obtain the monitoring screen images of multiple cameras deployed in the warehouse, perform boundary detection on the ground part in the monitoring screen images to obtain the ground area; perform cross-detection on the ground areas in each monitoring screen image to generate a non-blind area placement area map; A goods placement blind area detection module, configured to place goods according to the non-blind area placement area map when new goods are put into the warehouse, and perform monitoring blind area detection on the placed goods. If a new monitoring blind area is detected, an alarm is issued, and the recommended position for adding a camera is recommended to eliminate the new monitoring blind area.

Images