Bird intrusion detection method for tower scene

Abstract

The present invention discloses a method for detecting bird intrusion in a tower scene, comprising the steps of: obtaining monitoring images of a target area in several frames, performing target detection using a bird target monitoring model, and if a bird is detected in a certain frame, obtaining M consecutive monitoring images after the frame; judging whether a bird is detected in each frame of the M-frame monitoring images, and whether the size of the candidate detection frame of the bird target in each frame is greater than a preset pixel value; judging whether the area change rate of the candidate detection frame of the bird target in all two adjacent frames meets a preset first judgment threshold; judging whether the center point position change rate of the candidate detection frame of the bird target in all two adjacent frames meets a preset second judgment threshold, and if so, determining that there is bird intrusion in the current target area. The present invention effectively reduces the problems of missed detection and false detection caused by single-frame detection, significantly improves the accuracy of detection, and ensures the stability and consistency of the detection system in an all-weather environment.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a flying bird intrusion detection method for an iron tower scene. Background Art

[0002] In critical infrastructure such as power towers, preventing bird intrusion is crucial to the safety and stable operation of equipment. Therefore, it is particularly necessary to develop timely and accurate detection methods. However, existing bird detection technologies still face many challenges in low light and complex background environments.

[0003] Currently, object detection technology based on computer vision and artificial intelligence has been applied to the field of bird detection. The common method is to use models such as convolutional neural networks (CNN) to process images obtained by surveillance cameras to identify birds. However, these methods still have the following major shortcomings:

[0004] 1. Adaptability issues for all-weather monitoring: The monitoring system needs to operate at different times during the day, evening, and night. The color images acquired during the day and evening have different visual characteristics from the infrared grayscale images acquired at night. Especially at night, due to the fusion of the color of the bird's feathers with the background, the image contrast is low and the missed detection rate is significantly increased. The current solution is usually to build detection systems for color images and grayscale images separately, but this increases costs and brings complexity to system coordination.

[0005] 2. False detection in complex environments: In the wild, the size and color of flying birds may be similar to other objects (such as flying insects, balloons, or distant airplanes). Especially in complex backgrounds and low light conditions, neural network models are prone to misidentifying non-flying bird targets as flying birds. This increase in false detection rate significantly reduces the reliability of the detection system, especially in dynamic or irregular backgrounds.

[0006] 3. Insufficient accuracy of dynamic target detection: Existing technologies mostly rely on single-frame images for target detection, which cannot fully utilize the continuous motion information of flying birds, resulting in low accuracy in dynamic target recognition. Under various environmental interferences, key information such as the movement trajectory and speed changes of flying birds cannot be fully analyzed, which exacerbates missed detections and false detections. Summary of the invention

[0007] In view of the shortcomings of the above-mentioned prior art, the present invention aims to provide a method for detecting birds based on multi-frame judgment and multi-feature matching. The method combines the dynamic information of multi-frame images with the bird feature matching technology, effectively improving the accuracy of bird detection in complex scenes, especially in the application of night grayscale images. By utilizing the dynamic information in continuous frames, the system can more accurately identify birds, reduce the occurrence of missed detection and false detection, and improve the reliability and practicality of the overall detection.

[0008] In order to achieve the above-mentioned invention object, the technical solution provided by the present invention includes:

[0009] The bird intrusion detection method for the tower scene includes the following steps:

[0010] S1. Acquire several frames of monitoring images of the target area, and use the trained bird target monitoring model to perform target detection on the monitoring images. If a bird is detected in a certain frame, acquire M consecutive monitoring images after the frame.

[0011] S2, determining whether a flying bird is detected in each of the M frames of monitoring images, and whether the size of the candidate detection frame of the flying bird target in each frame is greater than a preset pixel value, if not, determining that there is no flying bird in the current target area, if yes, proceeding to the next step;

[0012] S3, determining whether the area change rate of the candidate detection frame of the flying bird target in all two adjacent frames meets the preset first determination threshold, if not, it is determined that there is no flying bird in the current target area, if yes, proceed to the next step;

[0013] S4. Determine whether the center point position change rate of the candidate detection frame of the flying bird target in all two adjacent frames meets the preset second determination threshold. If not, it is determined that there are no flying birds in the current target area. If so, it is determined that there are flying birds invading the current target area.

[0014] Preferably, the monitoring images include color images acquired during the day and evening time periods, and grayscale images acquired during the night time period.

[0015] Preferably, in step S1, the confidence threshold for performing target detection on the monitoring image using the trained flying bird target monitoring model is 0.15-0.2.

[0016] Preferably, the method for calculating the area change rate of the candidate detection frames of the flying bird target in two adjacent frames in step S3 includes:

[0017] The widths w1 and w2 and the heights h1 and h2 of the candidate detection frames of the flying bird target in two adjacent frames are obtained respectively, and the area change rate AR is calculated by the following formula: .

[0018] Preferably, the method for calculating the center point position change rate of the candidate bird target detection frame in two adjacent frames in step S4 includes:

[0019] The center point positions (x1, y1) and (x2, y2) of the candidate detection frames of the flying bird target in two adjacent frames are obtained respectively, and the center point position change rate PR is calculated by the following formula: ; Among them, w1 and w2 are the widths of the candidate detection boxes of the flying bird target in two adjacent frames.

[0020] Preferably, the method further comprises the steps of:

[0021] S5, for the continuous M frames of monitoring images in which it is determined that there is bird intrusion in the current target area, extract the bird features in each frame and calculate the average feature vector;

[0022] S6. Calculate the similarity between the average feature vector and the reference feature vector to determine whether the similarity is greater than a preset threshold value of 0.8. If so, it is determined that there is bird invasion in the current target area. If not, it is determined that there are no birds in the current target area. The reference feature vector includes several feature vectors extracted based on the bird image.

[0023] Preferably, the reference feature vectors include: feature vectors for at least one bird species commonly found in the region where the target area is located in three time periods: daytime, evening and night.

[0024] Preferably, the feature vector includes: a bird's perspective feature vector of the front, side and back features of the bird.

[0025] Beneficial effects:

[0026] 1. Improved detection accuracy: The present invention introduces a continuous frame detection mechanism and combines the area change rate and center point position change rate of the flying bird target to make a comprehensive judgment on the movement trajectory of the flying bird, effectively reducing the problems of missed detection and false detection caused by single-frame detection, especially in dynamic scenes, and significantly improving the detection accuracy.

[0027] 2. Adapt to all-weather monitoring needs: The present invention proposes a unified processing solution for color images and grayscale images based on the image characteristics under different lighting conditions during the day, evening and night, avoiding the high cost problem of separately building detection systems in the prior art and ensuring the stability and consistency of the detection system in an all-weather environment.

[0028] 3. Reduce false alarm rate: By analyzing the size, position, change rate and other multi-dimensional information of the candidate detection frame of the bird, the present invention can effectively distinguish between flying birds and other similar objects (such as flying insects, balloons, etc.), greatly reducing the false alarm rate in complex backgrounds and improving the reliability and practicality of the system.

[0029] 4. Efficient identification of bird intrusions in various environments: The present invention extracts multi-dimensional features of birds (such as the front, side, back features of the bird, as well as the head and body features of the bird) and compares them with preset reference feature vectors. This enables accurate identification of birds in different perspectives and flight states, thereby enhancing the system's adaptability in diverse scenarios.

[0030] 5. System cost optimization: Compared with the independent detection systems in the prior art that are respectively for different time periods and environmental conditions, the present invention provides a unified detection model, which reduces equipment and maintenance costs, while improving the efficiency and coordination of system operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 A schematic flow chart of a bird intrusion detection method for a tower scene provided in a preferred embodiment of the present invention;

[0032] Figure 2 A schematic flow chart of a bird intrusion detection method for a tower scene provided in another preferred embodiment of the present invention;

[0033] Figure 3 It is a schematic diagram of the results of calculating the center point position change rate and area change rate in two adjacent frames by a bird intrusion detection method for an iron tower scene provided in another preferred embodiment of the present invention. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical solution and advantages of the present invention clearer, the present invention is further described below in conjunction with the accompanying drawings. In the description of the present invention, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inside", "outside" and the like indicate directions or positional relationships based on the directions or positional relationships shown in the accompanying drawings, which are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention.

[0035] Embodiment 1

[0036] like Figure 1 As shown, this embodiment provides a bird intrusion detection method for a tower scene, comprising the steps of:

[0037] S1. Acquire several frames of monitoring images of the target area, and use the trained flying bird target monitoring model to perform target detection on the monitoring images. If a flying bird is detected in a certain frame, acquire M consecutive monitoring images after the frame.

[0038] The target area refers to the key environment near the towers used for bird monitoring. The towers include but are not limited to communication towers and power towers. The images of the area are monitored to determine whether there is bird intrusion.

[0039] Since the activities of flying birds have all-weather dynamic characteristics, the monitoring images of the target area cover three time periods: daytime, evening and nighttime. The existing technology mainly obtains color images through optical cameras and binary images through infrared cameras. Color images contain more color information than binary images, so they have higher detection accuracy under normal lighting conditions. However, color images are easily interfered by background colors and light conditions in complex environments. For example, bright or colorful backgrounds may mislead the detection system and increase the false detection rate. Especially at night, the feathers of flying birds are often gray or black, which are difficult to distinguish from the surrounding environment, resulting in greater color interference.

[0040] Based on the above problems, the present invention proposes a universal detection solution. In a preferred embodiment, color images are acquired during the day and evening time periods, and grayscale images are acquired at night. Preferably, an image acquisition device with an infrared fill light function can be used to acquire the above two images.

[0041] In some special cases, such as at dusk and at night, due to the low contrast between the flying birds and the background in the image, existing target detection algorithms (such as the YOLOv9 model based on convolutional neural networks) have certain limitations in detecting flying birds, especially in low-contrast environments, with a high rate of missed detection. Taking the YOLOv9 target detection model as an example, its default confidence threshold is 0.25. In this case, although the detection accuracy of the model is high, the missed detection phenomenon is still serious due to the insufficient contrast between the flying birds and the background in the image.

[0042] In order to improve the sensitivity of detection and reduce missed detections, the present invention, in some preferred embodiments, appropriately lowers the confidence threshold of the detection model and sets it to between 0.15 and 0.2. By lowering the confidence threshold, the system can capture more candidate frames of flying bird targets, expand the detection range, and significantly reduce the missed detection rate of flying birds. However, this processing method also introduces a new problem, that is, while obtaining more candidate detection frames, a large number of non-flying bird candidate frames may be generated, resulting in an increase in the false detection rate. To address this problem, the present invention designs a subsequent multi-step detection method.

[0043] It should be noted that if a flying bird does exist in the monitoring area, it will be detected in multiple consecutive frames of images because it is in a continuous moving state. The present invention introduces a multi-frame judgment mechanism, and further verifies the detection results by setting a threshold M for multi-frame judgment. Specifically, when a flying bird target is detected in a certain frame, the subsequent M-frame continuous images will be subjected to target detection and analysis to ensure that the target is an actual flying bird, rather than an interference object or a misdetected object that appears accidentally. Through the multi-frame judgment mechanism, the present invention can effectively distinguish between a real flying bird target and a non-flying bird target that appears accidentally in a single-frame image. In the continuous M-frame images, if a flying bird target can be detected in each frame, and the detection frame feature meets the set requirements, it can be confirmed that the target is a real flying bird. If a flying bird is detected only in a certain frame, and continuous detection is not maintained in subsequent frames, or the detection frame feature of the detection result shows that the target does not meet the moving characteristics of a flying bird, then the previously detected target may be an interference target such as an airplane, an insect, and a balloon, and the possibility that the target is a flying bird can be ruled out. By setting the multi-frame judgment threshold M, the present invention effectively reduces the possibility of occasional false detections, while ensuring that actual flying bird targets can be robustly detected. This technical means not only reduces the false detection rate, but also ensures continuous tracking and high-precision detection of the target.

[0044] The flying bird target monitoring model can be selected or constructed by those skilled in the art based on the commonly used target detection model in the prior art, and its training method can also be carried out according to the existing standard method. Since this part is not the focus of the present invention, it will not be described in detail here.

[0045] S2. Determine whether a flying bird is detected in each of the M frames of monitoring images, and whether the size of the candidate detection frame of the flying bird target in each frame is greater than a preset pixel value. If not, it is determined that there is no flying bird in the current target area. If so, proceed to the next step.

[0046] Those skilled in the art can understand that, since there is a certain spatial distance between the image acquisition device and the tower, in addition to flying birds, other flying interferences such as insects, airplanes, balloons, etc. may also appear in the monitoring area. When these interferences are at a long distance, although they are within the monitoring range, they will not actually pose a threat to the safety of the tower; similarly, when insects fly by at a relatively close distance from the image acquisition device, they may also cause system misdetection. The common feature of the above interferences is that they occupy a small proportion of pixels in the monitoring image. To this end, the present invention sets a preset pixel value, and determines the target with a pixel proportion less than the preset value as a flying interference at a long distance or a flying insect at a short distance, thereby effectively filtering out such irrelevant interference. The setting of the preset pixel value is based on a comprehensive consideration of multiple factors, including the distance between the image acquisition device and the tower, the resolution and characteristics of the camera device, and the bird species, flight altitude and habits in the target area. These factors determine the size of the flying bird in the image at different distances. For different tower environments and equipment configurations, the adjustment of the preset pixel value can further optimize the detection effect, filter out irrelevant interferences, and ensure that the system focuses on monitoring actual threats. It should be understood that the specific setting method of the preset pixel value is not the core content of the present invention, and its purpose is to improve the detection accuracy and robustness of the system. Therefore, the setting of the preset pixel value can be flexibly adjusted according to the specific application environment and device characteristics, and will not be repeated here. In some preferred embodiments, the preset pixel value can be set to 100.

[0047] In the process of bird intrusion detection, since the flying angle of the bird relative to the image acquisition device is arbitrary, its motion trajectory has a high degree of uncertainty. This brings challenges to the detection of bird targets. In order to ensure accurate identification of bird intrusion behavior, the present invention fully considers two typical flight situations: one is that the bird flies horizontally parallel to the field of view of the image acquisition device, and the other is that the bird flies toward the image acquisition device.

[0048] In the first case, the flying direction of the bird is parallel to the field of view of the image acquisition device, and the position of the bird in the image will change significantly, while the size of the candidate detection frame of the bird target remains almost unchanged. The reason is that the actual distance of the bird within the field of view does not change significantly, so the projected area in the image is relatively stable and the size changes slightly, but due to the large lateral movement of the bird, the displacement of its position in the image is more obvious.

[0049] In the second case, the flying direction of the bird is toward the image acquisition device. At this time, the position of the bird target candidate detection frame changes little because the relative position of the bird in the image remains basically fixed. However, as the bird gradually approaches the image acquisition device, the size of its target candidate detection frame will increase significantly. This is because the distance between the bird and the image acquisition device continues to shorten, causing its projection area in the image to expand rapidly.

[0050] In addition to the two special cases mentioned above, the flight trajectory of a bird usually presents complexity at any angle, that is, the bird may move sideways, or fly toward or away from the image acquisition device. In this case, the position and size of the candidate detection frame of the bird target will change at the same time. As the bird moves in space, its relative position and size in the image will change. Therefore, the system needs to consider both position changes and size changes at the same time to more accurately capture the flight trajectory of the bird and identify its intrusion behavior.

[0051] Based on this observation, the present invention designs a dual judgment mechanism for the candidate detection frame of the flying bird target, which analyzes the size change rate and position change rate of the detection frame respectively. This mechanism ensures that the system can accurately identify and judge the intrusion behavior of the flying bird under different flight trajectories. Regardless of whether the flying bird is flying parallel to the field of view, flying toward or away from the image acquisition device, or flying at any other angle, the change in size or position provides a reliable basis for judgment for the system. Furthermore, due to the physiological characteristics of the flying bird, it will not remain stationary in the air during the flight. Therefore, in any two consecutive frames, the size or position of the candidate detection frame of the flying bird target will inevitably change. This feature makes it possible to more accurately track the dynamics of the flying bird through the analysis of the size and position changes of consecutive frames, avoid false detection and missed detection problems, and thus improve the detection accuracy and robustness of the system. The specific steps are as follows:

[0052] S3, such as Figure 3 As shown, it is determined whether the area change rate of the candidate detection frame of the flying bird target in all two adjacent frames meets the preset first judgment threshold. If not, it is determined that there is no flying bird in the current target area. If so, proceed to the next step.

[0053] This step can effectively filter out some non-threatening flying objects by analyzing the area changes of the candidate detection frames of the bird targets. For example, the area change rate of distant objects (such as airplanes, balloons) or insects flying by at close range is usually inconsistent with the real bird invasion, so these irrelevant targets can be excluded through this step. This is because birds usually constantly change the distance between them and the camera during flight, especially when the birds approach or move away from the image acquisition device, the detection frame area of ​​the target in the image will change significantly. Compared with stationary or distant objects, bird invasions show more obvious dynamic change characteristics, which can be captured by the area change rate. It should be understood that the area change rate refers to the degree of change in the area of ​​the candidate detection frame of the bird target in two adjacent frames in the two frames of images. The solution method can be the commonly used area ratio method. A specific example is given in some preferred embodiments, specifically including:

[0054] The widths w1 and w2 and the heights h1 and h2 of the candidate detection frames of the flying bird target in two adjacent frames are obtained respectively, and the area change rate AR is calculated by the following formula: .

[0055] The setting of the first judgment threshold needs to comprehensively consider the dynamic characteristics of the area change rate of the candidate detection frame of the bird target under different distances, flight speeds, environmental complexity and other factors, to ensure that the threshold can accurately distinguish between real bird intrusions and other irrelevant targets. Those skilled in the art can use statistical analysis methods to determine the specific value of the threshold. Specifically, it can be optimized through data collection, statistical analysis, experimental verification and dynamic adjustment mechanism to ensure that the threshold can accurately judge bird intrusion behavior in different scenarios and environments, and effectively distinguish other non-threatening targets. There are different first judgment thresholds for monitoring scenarios in different environments, so the present invention does not intend to further limit this value. In some preferred embodiments, the first judgment threshold can be set to 0-0.2.

[0056] S4, such as Figure 3 As shown, it is determined whether the center point position change rate of the bird target candidate detection frame in all two adjacent frames meets the preset second judgment threshold. If not, it is determined that there are no flying birds in the current target area. If so, it is determined that there are flying birds invading the current target area.

[0057] This step determines whether the bird is moving in the monitoring area by analyzing the rate of change of the center point position of the candidate detection frame of the bird target. This is crucial for accurately judging the intrusion behavior of the bird, because the position change of the bird during flight is a key clue. By analyzing the change in the center point position, the bird can be distinguished from other interference objects that may be stationary or slowly moving, such as suspended objects, fixed background objects in the distance, etc. The bird may have a variety of movement trajectories during flight, including straight flight, circling, and sharp turns. Compared with only judging the area change, the position change can provide more information about the direction of the bird's movement. In some cases, the area change of the bird may be small, but the position change is very obvious. By analyzing the rate of change of the center point position, the flight dynamics of the bird can be captured more comprehensively. It should be understood that the center point position change rate refers to the degree of change in the displacement of the center point of the candidate detection frame of the bird target in two adjacent frames, and its solution method can be a displacement determination method. In some preferred embodiments, considering further excluding the size difference or target movement size change difference between different targets, a method for calculating the center point position change rate is given, including:

[0058] The center point positions (x1, y1) and (x2, y2) of the candidate detection frames of the flying bird target in two adjacent frames are obtained respectively, and the center point position change rate PR is calculated by the following formula: ; Among them, w1 and w2 are the widths of the candidate detection boxes of the flying bird target in two adjacent frames.

[0059] The second judgment threshold is set mainly to ensure that the system can accurately distinguish between real bird intrusions and other non-threatening targets under various flight trajectories. The setting method is mainly based on the analysis of the bird's motion trajectory, combined with the bird's flight speed, direction and the displacement data of the interference object, through experimental verification and scene adaptability adjustment, and finally determine a suitable center point position change rate threshold. The setting of this threshold ensures that the system can accurately judge the movement state of the bird under various complex flight paths, and effectively filter out non-threatening targets with small position changes. Similar to the setting of the first judgment threshold, there are different second judgment thresholds for monitoring scenes in different environments, so the present invention does not intend to further limit this value. In some preferred embodiments, the second judgment threshold can be set to 0-3.

[0060] It should be noted that the present invention verifies the motion state of the bird from two different dimensions, namely the area change rate of the bird target and the center point position change rate, through the setting of steps S3 and S4, which means that both the area and position of the bird in the detection frame have changed in accordance with the preset threshold. The simultaneous establishment of these two independent factors indicates that the bird is not only approaching the monitoring area, but is also moving or has obvious trajectory changes, thereby greatly improving the accuracy of bird intrusion judgment. Through double verification, the system can more confidently confirm the bird intrusion behavior and reduce the misjudgment that may be caused by single-dimensional judgment.

[0061] Embodiment 2

[0062] like Figure 2 As shown, this embodiment is developed on the basis of the above-mentioned embodiment 1. This embodiment considers providing an additional verification step to further confirm the accuracy of the flying bird target. Its purpose is to enhance the judgment of the flying bird intrusion behavior in complex scenes or uncertain situations through the matching of flying bird features, thereby greatly reducing the phenomenon of false detection and missed detection.

[0063] The bird intrusion detection method for the tower scene also includes the steps of:

[0064] S5, for the continuous M frames of monitoring images in which it is determined that there is bird intrusion in the current target area, extract the bird features in each frame and calculate the average feature vector;

[0065] S6. Calculate the similarity between the average feature vector and the reference feature vector to determine whether the similarity is greater than a preset threshold value of 0.8. If so, it is determined that there is bird invasion in the current target area. If not, it is determined that there are no birds in the current target area. The reference feature vector includes several feature vectors extracted based on the bird image.

[0066] Specifically, for M consecutive monitoring images containing at least M candidate detection frames of flying bird targets, feature extraction is performed on each candidate detection frame of the flying bird target. At this time, each feature is a one-dimensional vector with 1024 dimensions. The average feature vector is calculated for these M flying bird features. The calculation formula is as follows:

[0067] ;

[0068] Where i is the eigenvector number, j is the element number in the eigenvector, represents the jth element of the i-th eigenvector, is the sum of the jth elements of the M eigenvectors, is the average eigenvector.

[0069] This step is based on the visual features of the bird (such as shape, color, posture, etc.), and complements the changes in area and position to ensure that the system can still provide reliable judgment basis when the judgment is unclear.

[0070] It should be noted that the reference feature vector refers to the preset feature data used to compare the detection object (i.e. the detected flying bird target) for similarity judgment. These feature vectors are the flying bird feature data extracted from multiple different angles and parts based on the known flying bird images, and are stored in the system database as an important basis for judging whether the target in the current monitoring image is a flying bird. This is because the flying bird target candidate detection frame may be different perspectives and different parts of the flying bird. Matching these features can improve the accuracy.

[0071] In some preferred embodiments, the specific composition of the reference feature vector is given, and the feature vectors of at least one common bird species in the region where the target area is located in three time periods: daytime, evening and night. At this time, the main considerations are the differences in bird feature characteristics in different time periods and the diversity of birds in the areas where different towers are located. On the one hand, the visual features of birds will vary due to different lighting conditions during the day, evening and night. For example, during the day when the light is sufficient, the color, outline and other features of the birds are more obvious, while in the evening when the light is weakened or at night, the color of the birds may not be easy to distinguish, or even difficult to separate from the background. In this case, simply relying on the color or brightness changes of the image may not be enough to accurately identify the bird target. Therefore, it is necessary to extract the corresponding bird feature vector according to the specific lighting conditions in different time periods (daytime, evening, nighttime). This can ensure that the system can accurately match the bird features under all-weather monitoring conditions without being affected by light. On the other hand, the geographical environment of the area where the tower is located determines the diversity of bird species. For example, large birds such as eagles and storks may be common in some areas, while small birds are dominant in other areas. Different types of birds have obvious differences in appearance, flight mode, color, etc. Relying on a single flying bird model is difficult to cover these diverse features and is prone to false detection or missed detection.

[0072] It should be understood that the feature vector is a mathematical representation of important information such as the shape, color, texture, and flight posture of the bird, which can help the system identify the bird target more accurately and quickly. In some preferred embodiments, the feature vector includes: a bird perspective feature vector of the front, side, and back features of the bird.

[0073] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A bird intrusion detection method for a tower scene, characterized in that: Includes steps: S1. Acquire several frames of monitoring images of the target area, and use the trained bird target monitoring model to perform target detection on the monitoring images. If a bird is detected in a certain frame, acquire M consecutive monitoring images after the frame. S2, determining whether a flying bird is detected in each of the M frames of monitoring images, and whether the size of the candidate detection frame of the flying bird target in each frame is greater than a preset pixel value, if not, determining that there is no flying bird in the current target area, if yes, proceeding to the next step; S3, determining whether the area change rate of the candidate detection frame of the flying bird target in all two adjacent frames meets the preset first determination threshold, if not, it is determined that there is no flying bird in the current target area, if yes, proceed to the next step; S4, determining whether the center point position change rate of the candidate detection frame of the flying bird target in all two adjacent frames meets a preset second determination threshold, if not, determining that there is no flying bird in the current target area, and if so, determining that there is a flying bird intrusion in the current target area; The method for calculating the area change rate of the candidate detection frame of the flying bird target in two adjacent frames in step S3 includes: The widths w1 and w2 and the heights h1 and h2 of the candidate detection frames of the flying bird target in two adjacent frames are obtained respectively, and the area change rate AR is calculated by the following formula: ; The method for calculating the change rate of the center point position of the candidate detection frame of the flying bird target in two adjacent frames in step S4 includes: The center point positions (x1, y1) and (x2, y2) of the candidate detection frames of the flying bird target in two adjacent frames are obtained respectively, and the center point position change rate PR is calculated by the following formula: ; Among them, w1 and w2 are the widths of the candidate detection boxes of the flying bird target in two adjacent frames.

2. The bird intrusion detection method for a tower scene as claimed in claim 1, characterized in that: The monitoring images include color images acquired during the day and evening time periods, and grayscale images acquired during the night time period.

3. The bird intrusion detection method for a tower scene as claimed in claim 1, characterized in that: In the step S1, the confidence threshold for performing target detection on the monitoring image using the trained flying bird target monitoring model is 0.15-0.

2.

4. The bird intrusion detection method for a tower scene according to claim 1, characterized in that: Also includes the steps: S5, for the continuous M frames of monitoring images in which it is determined that there is bird intrusion in the current target area, extract the bird features in each frame and calculate the average feature vector; S6. Calculate the similarity between the average feature vector and the reference feature vector to determine whether the similarity is greater than a preset threshold value of 0.

8. If so, it is determined that there is bird invasion in the current target area. If not, it is determined that there are no birds in the current target area. The reference feature vector includes several feature vectors extracted based on the bird image.

5. The bird intrusion detection method for a tower scene as claimed in claim 4, characterized in that: The reference feature vectors include feature vectors of at least one bird species commonly found in the region where the target area is located in three time periods: daytime, evening and night.

6. The bird intrusion detection method for a tower scene as claimed in claim 4 or 5, characterized in that: The feature vectors include: bird perspective feature vectors of the front, side and back features of the bird.

Images