Dynamic region matching and strategy automatic association method based on camera PTZ value

Abstract

The invention provides a dynamic region matching and strategy automatic association method based on a camera PTZ value. The method comprises the following steps: acquiring an installation pose of a camera; calculating a three-dimensional coordinate set of an irradiation area of the camera according to the installation pose and the PTZ value of the camera; matching the three-dimensional coordinate set of the irradiation area with an irradiation scene three-dimensional model of the camera, and associating a recognition strategy of the camera according to a matching result; and when the PTZ value of the camera changes, calculating a matching result of the three-dimensional coordinate set of the updated irradiation area and the irradiation scene three-dimensional model of the camera by using the PTZ change value, and dynamically calling a corresponding identification strategy. According to the method, the irradiation area information of the camera is automatically updated according to the PTZ value of the camera, the identification strategy can be dynamically associated based on the scene three-dimensional modeling matching result, automatic scene adaptation is achieved, manual configuration or later adjustment of the camera monitoring identification strategy is not needed any more, and therefore the intelligent level and the resource utilization rate of the system can be improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent security technology, and particularly relates to a dynamic area matching and policy automatic association method based on the PTZ value of a camera. Background Art

[0002] In the technical implementation of an intelligent security monitoring system, the dynamic modeling of the camera irradiation area and the precise configuration of the recognition strategy are the core links to ensure the monitoring efficiency. The current mainstream systems generally complete these two key tasks through manual intervention, and there are significant efficiency bottlenecks and reliability hidden dangers in their technical implementation paths, which are specifically reflected in the inherent defects in three technical dimensions: hardware parameter calibration, dynamic perspective response strategy, and spatial mapping.

[0003] The essence of the camera irradiation area is the geometric mapping of the optical projection model in a three-dimensional scene, and its boundary is jointly determined by the three-dimensional coordinates of the physical installation pose of the camera, the optical axis direction, the internal parameters of the lens, the focal length distortion coefficient, and the PTZ dynamic adjustment parameters (horizontal rotation angle, vertical pitch angle, zoom ratio). In the traditional technical architecture, the modeling of the initial irradiation area relies on manual off-line calibration: technicians need to arrange calibration objects such as checkerboards in the monitoring scene, measure the installation height, horizontal offset distance, and optical axis pitch angle of the camera through a laser rangefinder, and combine the focal length and field of view angle parameter table provided by the lens manufacturer to manually draw the boundary of the irradiation area on the two-dimensional image plane in the monitoring management software (such as the vertex pixel coordinates of a rectangular warning area). This geometric modeling method based on empirical formulas has two sources of errors: one is the limitation of the manual measurement accuracy of physical parameters. The height measurement error is usually more than ±5 cm, and the horizontal inclination error exceeds ±2°, resulting in a 10% to 15% spatial deviation between the theoretically calculated irradiation area boundary and the actual optical projection range; the other is the influence of lens nonlinear distortion. In actual imaging, the radial distortion coefficient of a wide-angle lens is not effectively compensated, and the geometric distortion rate of the edge field of view can reach 5% to 8%, causing obvious geometric deformation of the regularly shaped irradiation area (such as a rectangle) configured manually at the image edge, directly affecting the positioning accuracy of subsequent target detection.

[0004] When the camera adjusts its viewing angle through the PTZ function, such as rotating horizontally by 30° and zooming the focal length to 200mm, complex transformations will occur in the spatial range of the irradiation area and its projection on the image plane. The existing system lacks the ability to automatically solve geometric transformations and cannot update the irradiation area in real time according to the PTZ parameters: technicians need to carry calibration equipment to the site, re-measure the installation parameters at the new viewing angle, and repeat the offline calibration process. The time-consuming for a single configuration usually exceeds 15 minutes. For large-scale security projects with more than 100 PTZ cameras deployed, such as the airport campus monitoring system, the number of viewing angle adjustments caused by equipment inspection scenario changes can reach 3,000 to 5,000 times per year. The resulting manual configuration cost accounts for 25% to 35% of the total operation and maintenance expenditure. More critically, the lag of manual configuration results in a policy blank period for the monitoring system for several hours after the viewing angle is adjusted. During this period, if events such as target intrusion occur, it is extremely easy to cause missed alarms or false alarms due to inaccurate boundaries of the irradiation area. Especially in scenarios with extremely high real-time requirements, such as guarding key perimeter protection areas, this delay may cause serious security hazards.

[0005] The configuration of the recognition strategy depends on the accurate modeling of the irradiation area, and there is a complex geometric coupling between the two in the spatial mapping relationship. In traditional systems, security managers need to manually delimit recognition areas on the two-dimensional monitoring screen, such as virtual warning lines and prohibited stay areas. These two-dimensional areas based on pixel coordinates need to be mapped to the three-dimensional scene space through the camera calibration parameters. However, the existing technology fails to establish an effective spatial coordinate conversion mechanism: on the one hand, the internal and external parameters of the camera, especially the distortion coefficient, are not fully incorporated into the mapping model, resulting in a significant increase in the conversion error between the two-dimensional pixel coordinates and the three-dimensional world coordinates as the field of view angle increases. The positioning deviation in the edge area can reach 0.5 to 1 meter; on the other hand, when the PTZ parameters change, the optical axis direction and the field of view range of the camera change, and the corresponding relationship between the original two-dimensional recognition area and the three-dimensional scene completely fails. The system cannot automatically trigger a policy update, and manual re-delimitation of the area on the new screen is required. This process not only consumes a lot of manpower, but also easily leads to inconsistent policy configurations due to subjective judgment differences of operators. For example, in the scenario of cross-camera linkage, if the spatial overlap of the recognition areas at different viewing angles is less than 70%, the target tracking trajectory will be broken, forming a monitoring blind spot.

[0006] From the in-depth logical analysis of technical implementation, the core defect of the traditional solution lies in the lack of dynamic geometric modeling capabilities and an automated mapping mechanism for the strategy space. Existing systems rely on static data calibrated offline and cannot real-time calculate the three-dimensional boundaries of the irradiated area after changes in PTZ parameters. At the same time, at the software architecture level, the configuration module of the recognition strategy and the camera parameter module are independent of each other, and a spatial coordinate conversion engine based on multi-view geometry has not been established, resulting in the mapping relationship between the two-dimensional image area and the three-dimensional monitoring space always relying on manual experience calibration. This limitation in the technical architecture makes it difficult for the system to meet the core requirements of intelligent security for real-time accuracy when facing complex scene changes in dynamic perspective adjustment.

[0007] In summary, the method of manually configuring the irradiated area and recognition strategy in traditional security monitoring systems is essentially a phased technical solution limited by the insufficient automation of geometric modeling and the lack of multi-source data fusion capabilities. The resulting inefficiencies and false alarm risks have become the key bottlenecks restricting the intelligent upgrade of security systems. Therefore, it is necessary to construct a dynamic irradiated area modeling algorithm based on real-time calculation of the internal and external parameters of the camera, combined with the automatic mapping technology of three-dimensional scene coordinates and two-dimensional image coordinates, to achieve intelligent adaptive configuration of recognition strategies and fundamentally improve the reliability and operation and maintenance efficiency of the monitoring system. Summary of the Invention

[0008] The present invention addresses the problems existing in the prior art and provides a method that is simple to implement, low in cost, and can automatically achieve camera irradiated area matching.

[0009] To achieve the above object, the technical solution adopted by the present invention is as follows: The present invention provides a method for dynamic area matching and automatic strategy association based on the PTZ value of a camera, which mainly includes the following steps: Obtain the installation pose of the camera; Calculate the three-dimensional coordinate set of the irradiated area of the camera according to the installation pose and the PTZ value of the camera; Match the three-dimensional coordinate set of the irradiated area with the three-dimensional model of the irradiated scene of the camera, and associate the recognition strategy of the camera according to the matching result; When the PTZ value of the camera changes, use the PTZ change value to calculate and update the matching result of the three-dimensional coordinate set of the irradiated area and the three-dimensional model of the irradiated scene of the camera, and dynamically call the corresponding recognition strategy.

[0010] Optionally, the modeling method of the three-dimensional model of the irradiated scene of the camera includes the following steps: After the camera is installed, capture the initial image set of the irradiated scene through the camera; Obtain a set of fixed-point images of the illumination scene of the camera through multi-point shooting by a drone; Based on the initial image set and the fixed-point image set, obtain a three-dimensional model of the illumination scene of the camera through a three-dimensional reconstruction method.

[0011] Optionally, the drone performs patrol along the edge in the illumination scene of the camera in a fixed-height mode, and sequentially takes scene pictures at multiple fixed points on the scene edge; The scene pictures form the fixed-point image set in the shooting order, and at least one scene picture contains the corresponding camera image.

[0012] Optionally, the three-dimensional reconstruction method includes the following steps: Perform image preprocessing on the initial image set and the fixed-point image set, and obtain a multi-view image set of the scene after merging; Establish a global coordinate system based on the installation position of the camera; Use the SIFT algorithm to extract image features from the multi-view image set, and sequentially match adjacent images based on the image features to obtain matching pairs; Based on the global coordinate system and the matching pairs, batch generate three-dimensional space points through the structure from motion technology framework; Establish a three-dimensional model of the illumination scene of the camera based on the three-dimensional space points.

[0013] Optionally, the image preprocessing includes geometric distortion correction and illumination equalization processing.

[0014] Optionally, calculating the three-dimensional coordinate set of the illumination area of the camera includes the following steps: Calculate the optical axis direction vector of the camera; Calculate the intersection point of the optical axis and the ground based on the optical axis direction vector; Calculate the boundary coordinates of the illumination area of the camera according to the internal parameters of the camera and the intersection point of the optical axis and the ground; Mark the illumination area of the camera through the boundary coordinates and obtain a three-dimensional coordinate set.

[0015] Optionally, assign a spatial position identifier to the three-dimensional model of the illumination scene, and establish a safety rule library based on the spatial position identifier; Establish an image recognition algorithm library, and construct an association rule between the safety rule library and the image recognition algorithm library based on the mapping method.

[0016] Optionally, the acquisition of the recognition strategy of the camera includes the following steps: Based on the spatial geometry algorithm, calculate the intersection of the irradiation area of the camera and the three-dimensional model of the irradiation scene of the camera; obtain the matching result based on the intersection; Obtain the spatial position identifier in the three-dimensional coordinate set of the irradiation area based on the matching result, and correspondingly obtain the matching entry in the security rule library; Extract the corresponding image recognition algorithm from the image recognition algorithm library based on the matching entry and the association rule; The corresponding image recognition algorithms are combined to form the recognition strategy of the camera.

[0017] Optionally, when the PTZ value of the camera changes, first compare the matching results before and after the change of the PTZ value; When the matching results before and after the change of the PTZ value are the same, do not re-invoke the recognition strategy; When the matching results before and after the change of the PTZ value are different, re-invoke the recognition strategy and replace the original recognition strategy.

[0018] Optionally, the following steps are further included: When the shooting target moves out of the irradiation area of the camera, judge whether target tracking is required based on preset conditions; if target tracking is required, trigger the drone to track the target, and select the corresponding image recognition strategy based on the images captured by the drone to recognize the area where the shooting target is located.

[0019] Compared with the prior art, the present invention has the following beneficial effects: The method of the present invention automatically updates the irradiation area information of the camera according to the PTZ value of the camera, and can dynamically associate the recognition strategy based on the matching result of the scene three-dimensional modeling, realizing automatic scene adaptation, and no longer requiring manual configuration or post-adjustment of the camera monitoring recognition strategy, thereby improving the intelligent level and resource utilization rate of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] 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 embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0021] Figure 1 It is the flowchart of the method in the specific embodiment of the present invention; Figure 2 It is the scene schematic diagram in the specific embodiment of the present invention.

[0022] In the figure: 1, PTZ camera, 2, monitoring scene, 3, water area. Detailed implementation manners

[0023] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are only a part rather than all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0024] It should be noted that similar reference numerals and letters denote similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0025] In the description of the present invention, "a plurality of" means two or more unless otherwise specifically defined.

[0026] It is worth noting that the methods used in the present invention are all conventional methods unless otherwise specified; the raw materials and devices used are all conventional commercially available products unless otherwise specified, and their sources are not specifically limited.

[0027] As Figure 1 shown, this embodiment provides a dynamic area matching and strategy automatic association method based on the PTZ value of a camera, which mainly includes the following steps: Combined with Figure 2 shown, at least one PTZ camera 1 (hereinafter referred to as the camera) is installed correspondingly in the monitoring scene 2 of each camera, which is used to obtain monitoring data, and the installation pose of the camera, that is, the position and direction of the camera, is obtained during the installation process. Among them, the position is reflected by the height H from the ground and the horizontal position of the camera; the direction is the installation direction of the camera base, which is reflected by a vector. Since the general installation position is horizontal installation, it is usually a polar coordinate value.

[0028] The PTZ value of the camera is a general term for describing three core parameters of the pan, tilt, and zoom of the monitoring camera, which directly affects the monitoring range, perspective adjustment, and picture clarity of the camera. Among them, pan: controls the horizontal rotation angle of the camera and determines the horizontal coverage range. Tilt: adjusts the vertical angle and affects the pitch monitoring range. Zoom: determines the degree of zooming in or out of the lens. The PTZ value can be specifically obtained through parameter settings at the system level.

[0029] By obtaining the above installation pose and PTZ value, the specific spatial position of the camera can be described from a spatial perspective.

[0030] Thus, calculate the three-dimensional coordinate set of the illumination area of the camera according to the installation pose and the PTZ values of the camera. Optionally, calculating the three-dimensional coordinate set of the illumination area of the camera includes the following steps: First, calculate the optical axis direction vector of the camera; the direction vector of the optical axis can be determined according to the installation pose and the PTZ values of the camera, specifically the Pan and Tilt of the pan-tilt head.

[0031] After that, calculate the intersection point of the optical axis and the ground based on the optical axis direction vector; given the installation height of the camera, use the projection method to solve the intersection point (x0, y0) of the extension line of the optical axis and the ground, so as to obtain the required intersection point of the optical axis and the ground.

[0032] According to the Zoom in the camera internal parameters and the intersection point of the optical axis and the ground, calculate the boundary coordinates of the illumination area of the camera through geometric algorithms, such as Figure 2 Some of the boundary points represented by the black dots in the figure.

[0033] Finally, mark the illumination area of the camera through the boundary coordinates and obtain the three-dimensional coordinate set.

[0034] Match the three-dimensional coordinate set of the illumination area with the three-dimensional model of the illumination scene of the camera, and associate the recognition strategy of the camera according to the matching result. Optionally, in this embodiment, it is also necessary to pre-obtain the three-dimensional data of the monitoring scene 2, and thus perform corresponding modeling operations. There are many executable ways for the method of three-dimensional scene reconstruction in the prior art. Among them, for the three-dimensional modeling of simple scenes, it can be restored in the computer with reference to engineering drawings, such as flat sites, etc.; for complex scenes, especially those with changes over time and a large number of irregularly shaped devices and objects, etc., manual surveying and mapping, automated surveying and mapping, etc. can be used. Specifically, in automated surveying and mapping, a drone can be equipped with a radar or a binocular vision system to complete the surveying and mapping work during the inspection process, and then the computer three-dimensional reconstruction / restoration of the scene can be carried out by the surveying and mapping data. The advantage of this method is that it is technically mature and has high precision. However, it requires professional drones and auxiliary equipment, and the cost of separate purchase and configuration is high, and it requires professional personnel to operate. It is suitable for three-dimensional surveying and mapping of large natural scenes, such as canyons, hills, river channels, etc.; if it is leased, although the usage cost can be reduced, for complex and time-varying scenes, such as the construction stage of the scene, it will change frequently with the construction progress, not only the morphology, but also the entry and exit of a large number of large equipment. The above situation will lead to frequent re-measurement requirements, and thus the lease cost will also increase significantly and affect the overall work progress; especially, complex re-measurement logics need to be specified to adapt to the scene requirements, such as when to perform re-surveying and modeling, and how to select the recognition strategy of the camera, etc.

[0035] In summary, this embodiment also provides a method for modeling a three-dimensional model of the illumination scene of a camera, which includes the following steps: After the camera is installed, capture an initial image set of the illumination scene through the camera; that is, after the camera is installed, image capture of part of the scene is performed in the initial state, and the images are aggregated to form an initial image set.

[0036] After that, obtain a fixed-point image set of the illumination scene of the camera by multi-point shooting with a drone equipped with a conventional monocular camera; during this period, this embodiment needs to use a drone with a height-keeping function. Since most commercially available products have a height-keeping function, such as barometric height-keeping, ultrasonic height-keeping, laser height-keeping, GPS height-keeping or a combination thereof, etc., it can be selected according to the accuracy requirements, and this embodiment does not make specific restrictions. The drone performs patrol along the edge of the illumination scene of the camera in the height-keeping mode, and periodically takes scene pictures at multiple fixed points along the path; the scene pictures form a fixed-point image set according to the shooting order, and at least one scene picture contains the corresponding camera image. The camera image should be understood that during the fixed-point shooting of the drone, at least one of the images taken at at least one point contains an element of a complete camera. In this embodiment, the difference between the height-keeping flight route of the drone and the height H of the camera during fixed-point shooting is set to 10m, rather than patrolling at the same height. Such a design can form a certain parallax; then, circular shooting is performed with a fixed camera as the center or along the edge of the scene through a manually set GPS path, shooting at intervals of 15°-30° in the horizontal direction, and at the same time supplementing vertical downward shooting images to obtain multi-angle parallax data. When the drone performs interval shooting in the horizontal direction, the line connecting the geometric center of the drone and the geometric center of the scene or the geometric center of the camera installation position forms an included angle, and the included angle between adjacent lines is the included angle; further, the fixed-point shooting direction of the drone faces the horizontal geometric center of the scene; furthermore, the POS system (GPS+IMU) carried by the drone records the external parameters of each image in real time, such as three-dimensional position, pitch, yaw and roll angles, etc., and internal parameters, such as focal length, distortion coefficient, etc. The above content can be obtained in advance through factory calibration or offline checkerboard calibration.

[0037] Based on the initial image set and the fixed-point image set, obtain a three-dimensional model of the illumination scene of the camera through a three-dimensional reconstruction method. Among them, the three-dimensional reconstruction method specifically includes the following steps: Perform image preprocessing on the initial image set and the fixed-point image set. The image preprocessing includes geometric distortion correction and illumination equalization processing; specifically, use image processing libraries such as OpenCV, and based on the known internal parameters of the camera and the drone, perform geometric distortion correction on all images to eliminate the influence of lens distortion, such as radial distortion and tangential distortion. Perform illumination equalization processing on overexposed and underexposed images, such as histogram equalization, etc., to enhance the feature detectability of weak texture areas.

[0038] The pre - processed images are merged to obtain a multi - perspective image set of the scene; A global coordinate system is established based on the installation positions of the cameras; its internal parameters such as focal length, optical center coordinates, distortion coefficients, etc., and external parameters such as three - dimensional spatial position, horizontal and vertical orientation angles, etc. are obtained through checkerboard calibration or device parameters and used as the origin of the global reference coordinate system. Specifically, the installation position of the camera is taken as the origin of the world coordinate system, and the X - axis, Y - axis, and Z - axis are defined. The GPS coordinates during the UAV shooting are converted into the coordinates of the local coordinate system with the camera as the origin through a geographic coordinate conversion tool.

[0039] The SIFT algorithm is used to extract image features from the multi - perspective image set, extracting key points and descriptors from all images. Each image has about 5000 feature points, with the ability to resist scale and rotation changes. In the multi - perspective image set, the images are arranged in order. Specifically, starting from the images taken by the camera and following the fixed - point order of the UAV for subsequent sorting, a multi - perspective image set is formed. Then, adjacent images are matched based on image features in sequence. That is, through the FLANN fast matching algorithm, one - to - one feature matching is performed between the UAV images and the fixed - camera images, and then the RANSAC algorithm is used to eliminate the mismatched points, retaining more than 80% of the correct matching pairs to ensure accurate cross - device perspective association.

[0040] Based on the global coordinate system and the matching pairs, a batch of three - dimensional space points are generated through the structure - from - motion technology framework. Specifically, known external parameters of the cameras and the UAV are used directly for triangulation to generate the first batch of three - dimensional space points. Then, the bundle adjustment (BA) algorithm is used to optimize the UAV pose, only adjusting the rotation and translation parameters of the UAV to minimize the reprojection error of the three - dimensional points on the images and improve the global reconstruction accuracy. Thus, depth map estimation and point cloud generation are completed. A three - dimensional model of the camera's illumination scene is established based on the three - dimensional space points. Specifically, the TSDF (Truncated Signed Distance Function) fusion technology is used to merge all sparse point clouds into a dense point cloud, filtering out outlier noise points by setting a distance threshold to generate a smooth and continuous point cloud model. The Poisson reconstruction algorithm or the Ball Pivoting algorithm is used to convert the dense point cloud into a triangular mesh model, thereby obtaining a three - dimensional model of the camera's illumination scene.

[0041] In this embodiment, in order to be able to flexibly call the camera recognition strategy, spatial position identifiers are assigned in the three - dimensional model of the illumination scene, such as Figure 2As shown in the figure, a feasible method is automated area marking based on semantic segmentation. For example, water surface features are extracted from an image. If there is a water area 3, the spatial position range is marked and the label "water area" is added. Another example is that if there are flammable materials in area A, the spatial position range is marked and the label "flammable area" is added. Then, a safety rule library based on the spatial position identification is established according to the labels in the spatial position identification. For example, "water area" corresponds to no entry for personnel, and "flammable area" corresponds to no smoking and no open flames.

[0042] An image recognition algorithm library is established corresponding to the entries in the safety rule library. Among them, different image recognition algorithms can efficiently and quickly identify the corresponding image content. For example, the cigarette recognition algorithm trained based on the convolutional neural network (CNN) commonly used in the retail or security inspection field can be used for the detection of the "flammable area". Thus, an association rule between the safety rule library and the image recognition algorithm library is constructed based on the mapping method, forming a data chain such as "spatial position identification [flammable area] - safety rule library [no smoking, no open flames] - image recognition algorithm library [cigarette recognition algorithm, open flame detection algorithm]".

[0043] Therefore, the following steps are included in the acquisition of the recognition strategy of the camera: Based on the spatial geometry algorithm, calculate the intersection of the irradiation area of the camera and the three-dimensional model of the irradiation scene of the camera; obtain the matching result based on the intersection. Here, the matching result is reflected in which spatial position identifications are included in the irradiation area, and the corresponding matching entries in the safety rule library are obtained. Taking a specific example, when the camera is initially installed, through the matching result, it is known that the irradiation area covers the water area and area A, and the spatial position identifications [water area] and [flammable area] are obtained. Thus, the system retrieves the safety rule library and obtains the entries [no entry for personnel] and [no smoking, no open flames]. Then, based on this matching entry and the association rule, the corresponding image recognition algorithms [personnel detection algorithm, cigarette recognition algorithm, open flame detection algorithm] are extracted from the image recognition algorithm library; The corresponding three groups of image recognition algorithms are combined to form the recognition strategy of the camera, and an alarm message is sent when the corresponding phenomenon is detected.

[0044] When the PTZ value of the camera changes, use the PTZ change value to calculate and update the matching result of the three-dimensional coordinate set of the irradiation area and the three-dimensional model of the irradiation scene of the camera, and dynamically call the corresponding recognition strategy. Optionally, since the PTZ camera 1 can be adjusted according to the remote setting, when the PTZ value of the camera changes, first compare the matching results before and after the PTZ value change; When the matching results before and after the PTZ value change are the same, that is, the intersection of the irradiation area of the camera and the three-dimensional model of the irradiation scene of the camera contains the same spatial position identifications, there is no need to update the recognition strategy of the camera. Therefore, the system does not re-call the recognition strategy to reduce subsequent operation steps and save system resources; When the matching results before and after the change of the PTZ value are inconsistent, that is, when the intersection of the irradiation area of the camera and the three-dimensional model of the irradiation scene of the camera contains different spatial position identifiers, the security rule library - image recognition algorithm library is retrieved in sequence according to the matching results to obtain a new algorithm combination, and the recognition strategy called last time is replaced, so as to realize the dynamic calling of the algorithm with the change of the PTZ value, so as to achieve the matching of the automated algorithm and the irradiation area of the camera, save system resources, and no longer need to pre-store a large number of algorithms to detect the camera images in parallel.

[0045] Optionally, in the specific monitoring process of this embodiment, because there are certain blind spots in the scene due to factors such as ground elevation difference and equipment orientation, and some blind spots will be utilized by personnel to avoid monitoring. Therefore, considering that the aforementioned UAVs in the embodiment are not high-cost professional equipment, the method further includes the following steps: When the shooting target moves out of the irradiation area of the camera, it is judged whether target tracking is required based on preset conditions; specifically, since the irradiation area of this embodiment includes a three-dimensional space, there are two cases: one is that the target continuously moves out of the irradiation area boundary; the other is that the target disappears within the irradiation area, such as the target is not detected for consecutive N frames. The latter can be understood as the target entering the monitoring blind spot. Therefore, during the target tracking process, it is judged by the position comparison method. If the target disappearance position is within the irradiation area, target tracking is required and the UAV is triggered to track the target. The UAV converts the target disappearance position provided by the monitoring system into a natural geographical coordinate and automatically cruises to that location for multi-angle shooting. Based on the images taken by the UAV, the corresponding image recognition strategy is selected to identify the area where the shooting target is located. That is, after the system receives the images taken by the UAV, it calculates the information of the spatial position identifiers in the area taken by the UAV according to the image content, and retrieves the recognition strategy for recognition by referring to the strategy of the fixed camera. This will not be elaborated here.

[0046] Through the above UAV collaborative monitoring, blind spot-free monitoring can be realized, the mobility of the UAV can be reasonably utilized, the necessary deployment quantity can be guaranteed, the system resources can be reasonably utilized, and the scene monitoring cost can be reduced.

[0047] Finally, it should be noted that the above content is only used to illustrate the technical solution of the present invention, rather than a limitation on the protection scope of the present invention. Any simple modification or equivalent replacement of the technical solution of the present invention by those of ordinary skill in the art does not depart from the essence and scope of the technical solution of the present invention.

Claims

1. A method for automatically associating dynamic area matching and strategy based on camera PTZ value, characterized by: The steps include: Get the installation position of the camera; Calculating a three-dimensional coordinate set of the illumination area of ​​the camera according to the installation posture and the PTZ value of the camera; Matching the three-dimensional coordinate set of the illumination area with the three-dimensional model of the illumination scene of the camera, and associating the recognition strategy of the camera according to the matching result; When the PTZ value of the camera changes, the PTZ change value is used to calculate and update the matching result between the three-dimensional coordinate set of the illumination area and the three-dimensional model of the illumination scene of the camera, and the corresponding recognition strategy is dynamically called.

2. The method for automatically associating dynamic area matching and strategy based on camera PTZ value according to claim 1, characterized in that: The method for modeling the three-dimensional model of the scene illuminated by the camera comprises the following steps: After the camera is installed, an initial set of images of the illuminated scene is captured by the camera; Acquire a set of fixed-point images of the scene illuminated by the camera through multi-point shooting by a drone; Based on the initial image set and the fixed-point image set, a three-dimensional model of the scene illuminated by the camera is obtained through a three-dimensional reconstruction method.

3. The method for automatic association of dynamic area matching and strategy based on camera PTZ value according to claim 2, characterized in that: The drone inspects along the edge of the scene illuminated by the camera in a fixed-altitude mode, and sequentially captures scene images at multiple fixed points on the edge of the scene; The scene graphs form the fixed-point image set in a shooting order, and at least one scene graph contains a corresponding camera image.

4. The method for automatic association of dynamic area matching and strategy based on camera PTZ value according to claim 2 or 3, characterized in that: The three-dimensional reconstruction method comprises the following steps: Performing image preprocessing on the initial image set and the fixed-point image set, and obtaining a scene multi-view image set after merging; Establishing a global coordinate system based on the installation position of the camera; Extracting image features from the multi-view image set using a SIFT algorithm, and matching adjacent images based on the image features in turn to obtain matching pairs; Based on the global coordinate system and the matching pairs, batch-generating three-dimensional space points through a structure-from-motion technology framework; A three-dimensional model of the scene illuminated by the camera is established based on the three-dimensional space points.

5. The method for automatic association of dynamic area matching and strategy based on camera PTZ value according to claim 4, characterized in that: The image preprocessing includes geometric distortion correction and illumination equalization processing.

6. The method for automatic association of dynamic area matching and strategy based on camera PTZ value according to claim 1, characterized in that: Calculating the three-dimensional coordinate set of the illumination area of ​​the camera comprises the following steps: Calculate the optical axis direction vector of the camera; Calculating the intersection of the optical axis and the ground based on the optical axis direction vector; Calculating the boundary coordinates of the illumination area of ​​the camera according to the inner part of the camera and the intersection of the optical axis and the ground; The illumination area of ​​the camera is marked by the boundary coordinates, and a three-dimensional coordinate set is obtained.

7. The method for automatic association of dynamic area matching and strategy based on camera PTZ value according to claim 1, characterized in that: Assigning a spatial position identifier in the three-dimensional model of the irradiation scene, and establishing a safety rule library based on the spatial position identifier; An image recognition algorithm library is established, and association rules between the security rule library and the image recognition algorithm library are constructed based on a mapping method.

8. The method for automatic association of dynamic area matching and strategy based on camera PTZ value according to claim 7, characterized in that: The camera recognition strategy acquisition includes the following steps: Based on a spatial geometry algorithm, calculating the intersection of the illumination area of ​​the camera and the three-dimensional model of the illumination scene of the camera; and obtaining the matching result based on the intersection; Based on the matching result, a spatial position identifier in the three-dimensional coordinate set of the irradiated area is obtained, and a matching entry in the security rule library is correspondingly obtained; extracting a corresponding image recognition algorithm from the image recognition algorithm library based on the matching entry and the association rule; The corresponding image recognition algorithms are combined to form the recognition strategy of the camera.

9. The method for automatic association of dynamic area matching and strategy based on camera PTZ value according to claim 8, characterized in that: When the PTZ value of the camera changes, firstly, the matching results before and after the PTZ value change are compared; When the matching results before and after the PTZ value changes are consistent, the recognition strategy is not called again; When the matching results before and after the PTZ value changes are inconsistent, the recognition strategy is re-called and the original recognition strategy is replaced.

10. The method for automatic association of dynamic area matching and strategy based on camera PTZ value according to claim 1, characterized in that: The following steps are also included: When the target moves out of the illumination area of ​​the camera, it is determined whether target tracking is required based on preset conditions; if target tracking is required, the drone is triggered to track the target, and the corresponding image recognition strategy is selected based on the image taken by the drone to identify the area where the target is located.

Images