Glass curtain wall bird strike risk assessment method and device, computer equipment and readable storage medium

Abstract

The application relates to a glass curtain wall bird impact risk assessment method and device, computer equipment and a readable storage medium. The method comprises the following steps: acquiring an original image of a target area; detecting anti-collision facilities in the original image to obtain an anti-collision facility deployment condition of the target area; performing semantic recognition on the original image to obtain a detection result of a flight trajectory associated object of a to-be-detected flight target in the original image; performing depth estimation on the original image to obtain depth information of the original image; and determining an impact probability of the to-be-detected flight target on the target area according to at least one of the anti-collision facility deployment condition, the detection result of the flight trajectory associated object and the depth information. The method can improve the accuracy of glass curtain wall impact risk assessment.

Description

Technical Field

[0001] This application relates to the field of computer data processing technology, and in particular to a method, apparatus, computer equipment, and readable storage medium for assessing bird strike risk in glass curtain walls. Background Technology

[0002] With the acceleration of urbanization and the proliferation of high-rise buildings, glass curtain walls have been widely used in the facade design of modern buildings due to their aesthetic appeal and transparent visual effects. At the same time, bird conservation and the maintenance of urban biodiversity are receiving increasing attention, and the impact problems caused by glass curtain walls are gradually becoming a common concern in the fields of ecology, architecture, and urban planning.

[0003] When birds are in flight, the reflective or transparent properties of glass may prevent them from correctly identifying it as an obstacle, leading to collisions with glass curtain walls and resulting in bird injuries or fatalities. Therefore, an effective assessment of the impact risk of glass curtain walls is necessary to guide the deployment of anti-collision measures. Currently, the assessment of glass curtain wall impact risk mainly relies on manual on-site inspections and experience-based judgment, which suffers from low efficiency and accuracy, thus affecting the prevention of glass curtain wall impact accidents. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, apparatus, computer equipment, and readable storage medium for assessing bird strike risk of glass curtain walls that can improve the accuracy of the assessment, in order to address the above-mentioned technical problems.

[0005] In a first aspect, this application provides a method for assessing the risk of bird strikes to glass curtain walls, the method comprising:

[0006] Obtain the original image of the target region;

[0007] The original image is subjected to collision avoidance facility detection to obtain the deployment status of collision avoidance facilities in the target area;

[0008] Semantic recognition is performed on the original image to obtain the detection results of the flight trajectory associated objects of the target to be detected in the original image;

[0009] Depth estimation is performed on the original image to obtain the depth information of the original image;

[0010] The probability of the target flight object colliding with the target area is determined based on at least one of the following: the deployment status of the collision avoidance facilities, the detection results of the flight trajectory associated objects, and the depth information.

[0011] Secondly, this application also provides a bird strike risk assessment device for glass curtain walls, the device comprising:

[0012] The acquisition module is used to acquire the original image of the target area;

[0013] The detection module is used to detect anti-collision facilities in the original image to obtain the deployment status of anti-collision facilities in the target area;

[0014] The recognition module is used to perform semantic recognition on the original image to obtain the detection result of the flight trajectory associated object of the flight target to be detected in the original image;

[0015] An estimation module is used to perform depth estimation on the original image to obtain the depth information of the original image;

[0016] The determination module is used to determine the probability of the target flight target colliding with the target area based on at least one of the following: the deployment status of the collision avoidance facilities, the detection results of the flight trajectory associated objects, and the depth information.

[0017] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps included in any of the foregoing method embodiments.

[0018] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps included in any of the foregoing method embodiments.

[0019] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps included in any of the foregoing method embodiments.

[0020] The aforementioned methods, apparatus, computer equipment, computer-readable storage media, and computer program products for assessing bird strike risks to glass curtain walls, by simultaneously acquiring environmental images and performing collision avoidance facility detection, semantic recognition, and depth estimation, utilize the complementary information among these three methods. Collision avoidance facility detection provides the hard conditions for whether protection already exists, semantic recognition provides the basis for why birds and other flying targets are attracted, and depth estimation provides a physical scale for the degree of misjudgment. This enables a refined assessment from the presence or absence of risk to the extent of the risk, significantly improving the accuracy of assessment results for glass curtain wall impact risks. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a diagram illustrating the application environment of a bird strike risk assessment method for glass curtain walls in one embodiment.

[0023] Figure 2 This is a flowchart illustrating a bird strike risk assessment method for glass curtain walls in one embodiment.

[0024] Figure 3 This is a structural block diagram of a bird strike risk assessment device for a glass curtain wall in one embodiment;

[0025] Figure 4 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0027] Before describing this embodiment, the relevant technologies and their existing problems will be further explained:

[0028] In related technologies, assessors visually observe the environmental features surrounding the glass curtain wall (such as the presence of trees, sky reflections, etc.) and combine this with their personal experience to qualitatively judge the likelihood of impact in a specific area. Some improvement solutions also use a single sensor (such as a regular camera) to capture images of the curtain wall, and then manually inspect the images for visual elements that might attract birds, or use simple image processing algorithms to identify whether anti-impact stickers have been affixed to the glass. Furthermore, some solutions statistically analyze the number of impact events that have occurred in the history of a building, using historical data as a basis for future risk assessments.

[0029] The above-mentioned approach has at least the following problems: First, the assessment results are highly subjective and lack quantitative standards. Manual inspections and experience-based judgments rely heavily on the professional competence and personal experience of the assessors. Different personnel may have significantly different assessment results for the same area, making it difficult to form a unified and reproducible risk indicator, and also making it impossible to quantitatively compare the risks of different buildings or different areas.

[0030] Secondly, the assessment dimensions are too narrow, and information utilization is insufficient. Traditional methods typically focus only on a single factor (such as the presence of tree reflections), failing to comprehensively consider the combined impact of multiple dimensions on collision risk, including the deployment of collision avoidance facilities, environmental semantic content, and spatial depth information. Furthermore, they suffer from low assessment efficiency and are difficult to apply on a large scale. Manual inspections require significant manpower; for high-rise buildings or large curtain walls, on-site inspections are difficult, time-consuming, and costly, and high-altitude operations pose safety hazards. Methods based on historical statistics require long-term data accumulation, and historical data cannot reflect the dynamic changes in risk brought about by environmental changes (such as the growth of surrounding trees and the construction of new buildings).

[0031] Therefore, there is a need for a method for assessing the risk of bird strikes to glass curtain walls that can be performed efficiently and automatically, in order to solve the problems of traditional methods being subjective, having a single dimension, and being inefficient.

[0032] It should be noted that the terms "first," "second," etc., used in this application may be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application includes two or more. The term "and / or" used in this application includes one of the embodiments, or any combination of multiple embodiments.

[0033] The bird strike risk assessment method for glass curtain walls provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or located in the cloud or on other network servers. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, drones, low-altitude aircraft, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, projection devices, etc. Portable wearable devices can include smartwatches, smart bracelets, head-mounted devices, etc. Head-mounted devices can be virtual reality (VR) devices, augmented reality (AR) devices, smart glasses, etc. Server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0034] In one exemplary embodiment, such as Figure 2As shown, a method for assessing the risk of bird strikes to glass curtain walls is provided, and this method is applied to... Figure 1 Taking terminal 102 or server 104 as an example, the following steps are included:

[0035] Step 202: Obtain the original image of the target area.

[0036] The target area includes the spatial region on the glass curtain wall where the impact risk needs to be assessed. This spatial region can be divided according to different assessment granularities and application scenarios. For example, for a refined risk distribution assessment, the target area can be a single glass panel separated by window frames or sealant joints. Alternatively, for a rapid assessment of the overall building risk, the target area can also be a continuous facade area composed of multiple adjacent glass panels, or a selected specific area of ​​interest, such as a building corner or a large unobstructed curtain wall. The original images include two-dimensional image data containing the target area captured by an image acquisition device, such as an RGB camera mounted on a drone, a ground-mounted high-definition industrial camera, or a surveillance camera. The acquisition parameters can be configured to ensure that the original images have sufficient spatial resolution and dynamic range to support subsequent semantic recognition and depth estimation tasks.

[0037] This embodiment, by acquiring raw images of the target area, can transform the physical glass curtain wall of a building into a digital image that can be processed by the algorithm, providing a unified input benchmark for all subsequent calculation steps. Furthermore, by flexibly defining target areas of different granularities during the data acquisition phase, this embodiment can adapt to various application scenarios, ranging from partial renovations of individual buildings to impact risk surveys at the city scale.

[0038] Step 204: Perform collision avoidance facility detection on the original image to obtain the deployment status of collision avoidance facilities in the target area.

[0039] Collision avoidance facilities include physical structures or materials that alter the visual properties of glass, enabling birds to perceive the presence of obstacles. For example, collision avoidance facilities may include dotted or striped stickers adhered to the glass surface, special glass films with ultraviolet reflective or patterned coatings, and grilles or louvers installed on the exterior of curtain walls. The deployment status of collision avoidance facilities includes information determined based on detection results regarding whether the target area has been effectively covered by these facilities. This status information can be a binary classification label (such as "deployed" or "not deployed") or a continuous indicator representing coverage density or integrity.

[0040] In this embodiment, identifying whether a target area is already equipped with artificial intervention measures that can effectively reduce impact risk provides a basis for subsequent differentiated impact risk assessment. Specifically, considering that if an area with effective anti-collision facilities has been deployed, the physical properties of the area have been artificially altered, and the probability of birds visually recognizing it as a passable path will be significantly reduced. In this case, the inherent risk of the area (based on environmental semantics and depth) can be covered or ignored. Conversely, if no facilities have been deployed, further analysis of the area's own optical and environmental factors is required. Through this preliminary judgment, this embodiment can avoid unnecessary calculations on protected areas and concentrate assessment resources on areas more likely to have risks, thereby optimizing assessment efficiency.

[0041] Specifically, this embodiment employs deep learning technology for collision avoidance facility detection. First, a training dataset containing a large number of glass curtain wall image samples is pre-constructed. Each sample image is labeled either "with collision avoidance measures" (e.g., labeled 1) or "without collision avoidance measures" (e.g., labeled 0). During labeling, for glass blocks with dotted or striped collision avoidance stickers, even if the stickers are slightly damaged or faded, if the main structure is intact, it is still labeled as "present". Then, an image classification model is trained using this dataset. Each glass block region in the original image (which can be pre-extracted by the glass block detection model) is cropped, scaled to a preset pixel size, and input into the trained detection model. The model outputs a probability value between 0 and 1. If the output probability is greater than a preset probability threshold (e.g., 0.5), the target area is determined to have "collision avoidance facilities"; if the output probability is less than or equal to the preset probability threshold, it is determined to have "no collision avoidance facilities", thus obtaining the deployment status of the collision avoidance facilities.

[0042] Step 206: Perform semantic recognition on the original image to obtain the detection results of the flight trajectory associated objects of the target to be detected in the original image.

[0043] The detected flight targets include objects that require protection to prevent collisions with the target area, such as birds requiring collision protection. The associated objects of the flight trajectory include visual environmental elements that can influence the flight decisions of the target. For example, these associated objects may include trees reflected in glass (e.g., trees with fruit), the sky (e.g., a gradient background including clouds or sunset), and man-made structures (e.g., building windows, vents, etc., structures that might be misidentified as passageways). The detection results include structured data extracted from the original image using semantic recognition algorithms regarding the presence or absence of the aforementioned associated objects and their spatial distribution (e.g., pixel regions, contours, co-occurrence relationships).

[0044] Specifically, semantic segmentation techniques can be used to process the original image, segmenting corresponding regions in the image based on text prompts. Specifically, the text prompt list can be set to ["tree", "sky", "building"], corresponding to trees, sky, and buildings, respectively. The original image is input into the CLIPSeg model, which outputs three probability masks of the same size as the original image. Each mask represents the probability of each pixel belonging to its corresponding category. By thresholding the masks (e.g., setting a threshold of 0.5) and performing connected component analysis, the specific pixel location and region range of each semantic category in the image can be obtained. The above segmentation results, including "whether there is sky", "whether there are trees", "whether there are buildings", and their respective pixel proportions, together constitute the detection results of the flight trajectory-related objects.

[0045] Considering that collisions between flying targets and highly transparent, highly reflective areas such as glass curtain walls are not entirely accidental and random, but rather highly correlated with the environment reflected by the target area, let's take birds as an example. There are two common reasons for collisions: first, birds see the trees or sky reflected in the reflective glass and mistakenly believe the space behind the glass is accessible, thus flying towards it; second, birds see the space or indoor plants on the other side through the transparent glass and mistakenly believe they can pass through it to reach the other side. For example, a piece of glass facing the canopy of roadside trees and reflecting a large expanse of sky is far more likely to induce birds to fly towards it than a piece of glass reflecting the wall of an office building opposite.

[0046] In this embodiment, by performing semantic recognition on the original image and analyzing the environmental content presented by the glass curtain wall or reflected therein, the flight behavior knowledge of the flying target can be transformed into quantitative evaluation factors, thereby giving the collision probability estimate high interpretability and reproducibility.

[0047] Step 208: Perform depth estimation on the original image to obtain the depth information of the original image.

[0048] Depth information refers to the estimated distance from the real physical point corresponding to each pixel in the original image to the observation point (or directly to the surface of the glass curtain wall). This information can be presented in the form of a dense depth map, where the value of each pixel represents the relative or absolute depth at that location. For example, for a piece of glass reflecting the distant sky and nearby trees, its depth information will include a sky area with depth values ​​approaching infinity and a tree area with smaller depth values ​​(e.g., 5-15 meters). Considering that depth information is one of the main references for distance judgment of flying targets in images, it may cause flying targets to misjudge the distance, resulting in a collision with the target area. Specifically, the farther the distant scene (large depth value), the more likely the flying target is to believe that it can safely fly over it, and the greater the risk; the closer the near scene (small depth value), the easier it is for the flying target to perceive the presence of the obstacle, and the lower the risk.

[0049] For example, even if a piece of glass reflects trees, if the reflected image is very close to the glass (e.g., less than 1 meter), flying targets such as birds may still perceive the glass's presence through parallax or motion cues and avoid a collision. Conversely, if the reflected image has a very large depth of field (e.g., a wide-open field), flying targets such as birds are more likely to mistakenly believe that there is an unobstructed flight path ahead. Therefore, in this embodiment, depth information is extracted as a "weighted adjustment" for semantic incentives. Semantics provides the motivation to "fly there," while depth determines "whether one believes they can fly there."

[0050] Specifically, this embodiment uses a monocular depth estimation model to predict the depth of the original image. The original image is input into the monocular depth estimation model, which outputs a depth map of the same size as the original image. The value of each pixel represents the estimated depth at that location. For example, in an image containing buildings, sky, and trees in the foreground, the depth value of the sky region (at infinity) will be predicted by the model to be a large value (e.g., 80 to 100 meters), the depth value of the nearby trees will be smaller (e.g., 5 to 15 meters), and the depth value of the glass curtain wall itself depends on the shooting distance (e.g., 10 to 20 meters). The output depth map is the depth information. To facilitate subsequent calculations, necessary data cleaning can be performed on the depth values, such as removing outliers that exceed the effective range of the sensor (e.g., 0.5 to 100 meters).

[0051] Step 210: Determine the probability of the target object colliding with the target area based on at least one of the following: the deployment status of the collision avoidance facilities, the detection results of the flight trajectory associated objects, and the depth information.

[0052] The collision probability is a numerical value that characterizes the likelihood of a bird colliding with the target area. This value can be a dimensionless probability value between 0 and 1, or it can be converted into a risk level (such as high, medium, and low) or a risk score (such as 0-100 points) depending on the application requirements.

[0053] In this embodiment, the collision avoidance facility deployment, the detection results of objects associated with the flight trajectory, and the physical meaning and confidence level of depth information are selectively or weightedly fused to obtain the impact zone of the flying target on the target area. Specifically, considering that the physical presence of the collision avoidance facility can fundamentally and with a high probability disrupt the specular reflection and perspective properties of glass, the possibility of bird strikes has been verified to be negligible in actual engineering, regardless of how attractive the environment behind it may be. Therefore, preliminary risk diversion can be performed on the target area based on the deployment of the collision avoidance facility. Specifically, if the detection results show that effective collision avoidance facilities have been deployed in the target area, then based on the principle that facility effectiveness takes precedence over environmental factors, the impact probability of the area is directly determined to be the lowest value (e.g., 0).

[0054] Correspondingly, if the detection results show that no collision avoidance facilities have been deployed within the target area, the detection results of objects associated with the flight trajectory are further integrated with depth information. Specifically, firstly, based on the types of associated objects detected by semantic recognition and their co-occurrence patterns (e.g., whether "sky + trees" appear simultaneously, or only "buildings" appear), a semantic risk baseline is determined. This baseline reflects the inherent attractiveness of the current environment to birds, and its magnitude is positively correlated with the degree of influence of associated objects on the bird's flight trajectory. Then, depth information is used to correct this baseline.

[0055] Considering that the impact risk is positively correlated with the depth of the distant view and negatively correlated with the proximity of the near view, this embodiment statistically analyzes the spatial distribution in the depth map to extract a statistic that characterizes the spatial transparency or depth-of-field induction intensity. The semantic risk baseline value is then multiplied or weighted and summed with this depth statistic to obtain the impact probability of the target area, thereby ensuring that the final output probability value has both behavioral interpretability and physical spatial authenticity.

[0056] The bird strike risk assessment method for glass curtain walls provided in this embodiment simultaneously acquires environmental images and performs collision avoidance facility detection, semantic recognition, and depth estimation. By utilizing the complementary information among these three methods—collision avoidance facility detection providing the hard conditions of existing protection, semantic recognition providing the behavioral basis for why birds and other flying targets are attracted, and depth estimation providing the physical scale of the degree of misjudgment—it achieves a refined assessment from the absence of risk to the extent of risk, significantly improving the accuracy of the assessment results for the impact probability of glass curtain walls.

[0057] In some embodiments, the deployment of collision avoidance facilities includes the coverage of collision avoidance facilities within the target area; the detection result of the flight trajectory associated object includes the correlation degree between the flight trajectory associated object existing in the target area and the flight trajectory of the flight target to be detected; the depth information includes the depth value of each pixel in the target area;

[0058] The step of determining the probability of the detected flight target colliding with the target area based on at least one of the following: the deployment status of the collision avoidance facilities, the detection results of the flight trajectory associated objects, and the depth information, includes:

[0059] If the coverage indicates that the anti-collision facility is deployed within the target area, the collision probability is determined based on the coverage; wherein the collision probability is negatively correlated with the coverage of the anti-collision facility.

[0060] If the coverage indicates that no collision avoidance facility is deployed in the target area, the flight trajectory correlation degree and the depth value are weighted to obtain the collision probability; wherein, the collision probability is positively correlated with the flight trajectory correlation degree between the flight trajectory associated object and the flight trajectory of the target to be detected; and the collision probability is positively correlated with the average depth value of the pixels in the target area.

[0061] When collision avoidance facilities already exist in the target area, the probability of an impact is linearly reduced based on the completeness of their deployment, thus achieving a quantitative assessment that protection reduces risk. Coverage refers to the density or completeness of collision avoidance facilities within the target area. For example, coverage can be a continuous value between 0 and 1, where 1 represents complete coverage of the target area (e.g., covered with stickers), and 0 represents complete uncovering. This coverage can be further extracted from the collision avoidance facility detection results, for example, by calculating the proportion of detected sticker pixel area to the total pixel area of ​​the target area.

[0062] Considering that not all areas with collision protection facilities have absolutely zero risk, collision protection facilities such as stickers may partially detach due to aging, or may only be deployed in localized areas of glass (e.g., only dotted stickers on the edges). In such cases, the facilities can still reduce risk, but may not completely eliminate it. Therefore, this embodiment introduces a coverage parameter to distinguish the risk differences between full coverage and partial coverage, making the assessment results more refined. Specifically, if the coverage represents the presence of collision protection facilities in the target area (e.g., coverage greater than 0), the impact probability can be set to a value negatively correlated with the coverage, such as P = 1 – coverage, or other monotonically decreasing functions. When the coverage is 1 (full coverage), the impact probability is 0; when the coverage is 0.5 (half coverage), the impact probability is 0.5. Thus, this embodiment achieves differentiated risk assessment for areas with deployed facilities.

[0063] Correspondingly, when the target area has no physical protection, this embodiment integrates environmental semantic inducements (i.e., flight trajectory correlation) and spatial depth information (i.e., depth value) to quantify its inherent risk. Flight trajectory correlation refers to the induction intensity coefficient determined based on behavioral statistics between the detected flight trajectory-related objects and the detected flight target (such as birds). For example, different types of related objects have different correlation degrees: the co-occurrence of trees and the sky has the highest correlation degree (e.g., assigned a value of 1.0), followed by individual trees or the sky (e.g., 0.7), while individual buildings have a lower correlation degree (e.g., 0.6).

[0064] Considering the difference between areas with at least some collision avoidance facilities and areas without protection, the risk of unprotected areas is largely determined by their visual presentation. Specifically, semantic recognition detection results provide the motivation for a flying target to head towards the target area; the higher the correlation, the stronger the motivation. Depth information provides the flying target's judgment on whether it believes it can fly through the target area; the greater the average depth, the more likely the target area is to be misjudged as a safe passage by the flying target. Therefore, in this embodiment, the two are weighted to comprehensively reflect the influence of these two orthogonal factors. Specifically, the weighting can be performed in a multiplicative form: first, the depth values ​​of all pixels within the target area are statistically analyzed, for example, by calculating the average or the average after nonlinear transformation, to obtain a depth factor representing spatial transparency. Then, the correlation of the flight trajectory is multiplied by this depth factor as a weight to obtain the collision probability. Since the collision probability is positively correlated with the correlation (the higher the correlation, the greater the probability) and also positively correlated with the average depth value (the farther the depth of field, the greater the probability), this weighted result can truly reflect the difference between high-risk scenarios (high correlation + deep field) and low-risk scenarios (low correlation + close-up field). For example, for an unprotected glass window that reflects the sky and trees (correlation degree 1.0) and has a depth of over 10 meters (depth factor close to 1.0), its impact probability will approach 1.0; while for an unprotected glass window that only reflects buildings (correlation degree 0.6) and has a depth of only 2 meters (depth factor approximately 0.2), its impact probability is approximately 0.12. Thus, this embodiment achieves continuous and precise quantification of the risk in unprotected areas.

[0065] In some embodiments, if the coverage indicates that no collision avoidance facility is deployed within the target area, weighting the flight trajectory correlation and the depth value to obtain the collision probability includes:

[0066] The depth value of each pixel is normalized, and the average value of the normalized depth value of all pixels in the target region is calculated.

[0067] The collision probability is obtained by weighting the average value of the normalized depth value based on the correlation of the flight trajectory.

[0068] Normalization refers to the process of mapping the original depth value to a standard range (e.g., between 0 and 1) using a mathematical transformation function. This transforms the original depth value obtained from depth estimation (measured in meters, ranging from a few tenths of a meter to hundreds of meters) into a unified, bounded numerical interval, enabling meaningful multiplicative or additive fusion with the semantic risk coefficient. For example, normalization can employ nonlinear compression functions such as the hyperbolic tangent (tanh), sigmoid, or piecewise linear functions. Considering that the contribution of depth to impact risk does not increase indefinitely, beyond a certain distance (e.g., 30 to 50 meters), birds can no longer visually distinguish differences in distance, and the risk gain tends to saturate. Therefore, normalization ensures that the normalized depth values ​​of extremely distant pixels (e.g., at a depth of 100 meters) are not significantly different from those of relatively distant pixels (e.g., at a depth of 30 meters), which aligns with the visual physiological characteristics of birds.

[0069] Specifically, the normalization process can be performed using the following formula: Normalized depth = tanh(original depth / 10.0). Here, 10.0 represents an empirical scaling factor, ensuring that the typical depth range (0 to 30 meters) falls within the sensitive variation range of the tanh function; the output range of the tanh function is 0 to 1, outputting 0 when the depth is 0 and approaching 1 when the depth approaches infinity. For a piece of glass reflecting distant mountains (approximately 80 meters deep) and nearby trees (approximately 8 meters deep), the normalized depth of the sky region is approximately tanh(8) = 0.999, and the normalized depth of the tree region is approximately tanh(0.8) = 0.664.

[0070] After normalizing each pixel, the arithmetic mean of the normalized depth values ​​of all pixels within the target area is calculated. The specific formula is: Depth_avg = (Σ D_norm_i) / N, where D_norm_i is the normalized depth value of the i-th pixel, and N is the total number of pixels in the target area. This average value characterizes the overall average spatial transparency of the glass block. Compared to using the maximum or median, the average value more stably reflects the depth distribution characteristics of the entire glass area, avoiding excessive influence of individual extreme depth values ​​(such as a very small area of ​​sky producing a depth value close to 1) on the evaluation results.

[0071] Weighted processing refers to the operation of multiplying two factors (or using other monotonically increasing fusion functions) to multiplicatively fuse the inducing strength (relevance) of the semantic dimension with the transparency (average normalized depth) of the spatial dimension, outputting the final single glass block impact probability. The flight trajectory relevance is used as a weight multiplier because semantic inducement determines the upper limit of risk; that is, if the content reflected by the glass is unattractive to birds (relevance close to 0), then even with a great depth of field, the impact probability should be close to 0. Conversely, if the semantic inducement is extremely strong (relevance close to 1), the impact probability is mainly determined by the depth factor.

[0072] For example, the formula for calculating the impact probability can be: P = M × Depth_avg, where M is the flight trajectory correlation degree (or semantic risk coefficient), and Depth_avg is the average of the normalized depth values ​​calculated in step 220. In this multiplicative weighting method, semantic correlation degree and depth transparency are independent conditions that must be met simultaneously. A piece of glass that reflects the sky and trees (M=1.0) and has a very long depth of field (Depth_avg≈0.95) has an impact probability of approximately 0.95, which is considered high risk. A piece of glass that also reflects the sky and trees (M=1.0) but has a very short depth of field (Depth_avg≈0.05, for example, the glass is close to the opposite wall) has an impact probability of only 0.05, which is considered low risk. A piece of glass that only reflects buildings (M=0.6) and has a medium depth of field (Depth_avg approximately 0.5) has an impact probability of 0.3, which is considered low to medium risk. By using multiplicative weighting, this embodiment can accurately distinguish the risk differences in the above-mentioned different scenarios, providing a quantitative basis for prioritizing subsequent collision avoidance modifications.

[0073] It should be noted that the above weighting method is not limited to multiplication; a weighted summation can also be used: P = α * M + β * Depth_avg, where α and β are preset weight coefficients used to adjust the relative contributions of semantic and depth factors. Compared to weighted summation, the multiplicative form better reflects the logical relationship that both factors need to be present simultaneously, and P automatically approaches 0 when either M or Depth_avg is close to 0, which is more consistent with the physical mechanism of collision occurrence.

[0074] In some embodiments, the flight trajectory association object includes multiple optional association objects; determining the collision probability of the target flight target with the target area based on at least one of the collision avoidance facility deployment status, the detection result of the flight trajectory association object, and the depth information includes:

[0075] Determine the co-occurrence of the multiple optional associated objects within the target area; the co-occurrence includes the flight trajectory correlation degree corresponding to each co-occurring optional associated object.

[0076] The collision probability is determined based on the co-occurrence situation; wherein the collision probability is positively correlated with the maximum value or sum of the correlation degree of the flight trajectories of the co-occurring optional associated objects.

[0077] Among these, multiple optional associated objects refer to pre-defined visual environmental elements of different types that can influence the flight decisions of the target. For example, optional associated objects may include the sky, trees, and buildings. Co-occurrence refers to whether the above multiple optional associated objects appear simultaneously within the same target area, and specifically, what types of combinations they are. For example, co-occurrence may include the simultaneous appearance of the sky and trees, the simultaneous appearance of the sky and buildings, the simultaneous appearance of trees and buildings, the appearance of only a single object (such as only the sky, only trees, or only buildings), and the appearance of no associated objects (such as only other types).

[0078] Flight trajectory correlation refers to the induction strength coefficient determined based on avian behavioral statistics for each optional associated object. It is understandable that different flight trajectory associated objects have different flight trajectory correlation degrees. Taking birds as the flight target as an example, avian behavior shows that trees are generally more attractive to birds than buildings because trees provide ecological functions such as habitat and foraging. Specifically, a flight trajectory correlation degree can be pre-assigned to each optional associated object: 0.7 for trees, 0.7 for the sky, and 0.6 for buildings. The flight trajectory correlation degree for other unlisted objects (such as the ground, clouds, etc.) can be set to 0.5. It should be noted that the flight trajectory correlation degree values ​​corresponding to each flight trajectory associated object can be obtained through experimental statistics or literature review, and can be adaptively adjusted according to different regions and different types of flight targets (such as bird species).

[0079] Considering that impact risk is not determined by a single semantic factor, but rather by the combined effect of the overall environmental content reflected or transmitted through the glass, for example, a piece of glass that reflects both trees and sky has a much stronger inducing effect on birds than glass that only reflects trees or only sky. This is because trees provide destination signals, and the sky provides signals representing passable paths; the combination of these two creates a stronger inducement for misjudgment. By identifying co-occurrence scenarios, this embodiment can capture this synergistic enhancement effect caused by the coexistence of multiple flight path-related objects, thereby improving the accuracy of risk assessment.

[0080] Based on the co-occurrence situation, a comprehensive semantic risk coefficient is calculated by merging the flight trajectory correlation degrees of multiple associated objects. This coefficient will serve as the basis for subsequent weighted fusion with depth information. The collision probability is positively correlated with the maximum or sum of the flight trajectory correlation degrees of co-occurring optional associated objects. This means that when multiple associated objects co-occur, the semantic risk coefficient is higher than the flight trajectory correlation degree of a single object, thus reflecting the synergistic enhancement effect brought about by co-occurrence.

[0081] Specifically, when multiple possible associated objects appear simultaneously within the target area, the highest flight trajectory correlation is selected as the semantic risk coefficient for that area. For example, when both sky (flight trajectory correlation 0.7) and trees (flight trajectory correlation 0.7) are present, the maximum value of 0.7 is taken. Optionally, a merging strategy of taking the maximum value and adding a co-occurrence increment can be adopted. For example, based on engineering experience in impact risk assessment, the actual risk coefficients for different semantic combinations are as follows: when both sky and trees are present, the risk coefficient is 1.0; when both sky and buildings are present, the risk coefficient is 0.9; when both trees and buildings are present, the risk coefficient is 0.8; when only trees or only sky are present, the risk coefficient is 0.7; when only buildings are present, the risk coefficient is 0.6; and when only buildings are present, the risk coefficient is 0.5. The simple maximum value (0.7) or simple sum value (1.4) of the above empirical coefficients and the correlation between flight trajectory are not completely consistent. Instead, it reflects that "sky + trees" has the highest synergistic enhancement effect (1.0), followed by "sky + buildings" and "trees + buildings" (0.9 and 0.8 respectively).

[0082] Specifically, a mapping table can be pre-established to map co-occurrence conditions to semantic risk coefficients. This table records the risk coefficients corresponding to various possible co-occurrence combinations. The corresponding semantic risk coefficient M is obtained by looking up the table based on the co-occurrence condition. For example, the specific mapping rules for using the lookup table method can be as follows: if both sky and trees exist, then M = 1.0; if both sky and buildings exist, then M = 0.9; if both trees and buildings exist, then M = 0.8; if only trees exist (no sky, no buildings), then M = 0.7; if only sky exists (no trees, no buildings), then M = 0.7; if only buildings exist (no sky, no trees), then M = 0.6; if only other types exist (no sky, no trees, no buildings), then M = 0.5; if multiple conditions are met simultaneously (e.g., both sky, trees, and buildings exist), then the maximum value is taken, i.e., M = 1.0 (because the combination of sky and trees gives the highest value of 1.0).

[0083] After determining the semantic risk coefficient M, the collision probability is calculated by combining it with depth information. For example, the average depth value, Depth_avg, after normalization can be calculated, and then a multiplicative weighted average can be applied: P = M × Depth_avg. Since M is directly related to co-occurrence and its value is positively correlated with the correlation of co-occurring associated objects, the final collision probability is also positively correlated with the co-occurrence.

[0084] In some embodiments, the target flight to be detected includes a bird; after determining the probability of the target flight to be detected impacting the target area based on at least one of the collision avoidance facility deployment, the detection result of the flight trajectory associated object, and the depth information, the method further includes:

[0085] Obtain the flight speed and / or binocular field of view of the target to be detected;

[0086] The impact probability is corrected based on the flight speed and / or binocular field of view, wherein the corrected impact probability is positively correlated with the flight speed and negatively correlated with the binocular field of view.

[0087] Considering that different types of flying targets have different flight capabilities and visual perception structures, their ability to perceive and avoid glass curtain walls also varies significantly. For example, birds with high flight speeds may be more prone to collisions due to insufficient reaction time, while birds with lower flight speeds have more time to perceive and avoid obstacles. Therefore, flight speed refers to the horizontal flight rate of a specific bird under normal flight conditions (not diving, not taking off or landing), and its unit can be meters per second.

[0088] The binocular field of view (FAV) refers to the angle of overlap between the visual fields of a flying target's two eyes. This parameter affects the target's stereoscopic vision and distance perception accuracy. Specifically, the wider the FAV, the larger the overlap area between the two eyes, the stronger the target's ability to perceive the distance and depth of objects, i.e., the better its binocular stereoscopic vision. This allows the target to more accurately determine the position of the glass curtain wall and avoid it in advance, resulting in a lower probability of impact. Therefore, the corrected impact probability is positively correlated with the binocular FAV.

[0089] Correspondingly, the narrower the binocular field of view of a flying target, and the smaller the overlap area between the two eyes, the more the target relies on monocular vision, resulting in weaker distance perception and a higher probability of misjudging glass as a passable path, thus increasing the probability of collision. For example, a pigeon's binocular field of view is approximately 20° to 30°, indicating relatively limited stereoscopic vision; while an owl's binocular field of view can reach 50° to 70°, giving it better depth perception. The unit for binocular field of view can be degrees.

[0090] Specifically, a bird species parameter database can be pre-established, recording the species names, average flight speeds, and binocular field of view of common birds. For example, the database could include: Sparrow (Passermontanus), flight speed approximately 8 m / s, binocular field of view approximately 25°; Barn Swallow (Hirundo rustica), flight speed approximately 15 m / s, binocular field of view approximately 20°; Pigeon (Columba livia), flight speed approximately 10 m / s, binocular field of view approximately 30°; Magpie (Pica pica), flight speed approximately 12 m / s, binocular field of view approximately 28°. Specifically, the specific bird species being evaluated can be determined through image recognition or manual input, and then its corresponding flight speed and binocular field of view parameters can be retrieved from the database. Optionally, if species differentiation is not required, a representative average value (e.g., a flight speed of 11 m / s and a binocular field of view of 25°) can be used as a general correction parameter.

[0091] This embodiment incorporates the species characteristics of specific birds into the risk assessment model, enabling the final output impact probability to reflect the differentiated risk levels of different birds towards the same glass curtain wall. Correction refers to a multiplicative adjustment or functional transformation of the base impact probability based on flight speed and / or binocular field of view. The corrected impact probability is positively correlated with flight speed, meaning that the faster the bird flies, the higher its corrected impact probability (greater risk); the corrected impact probability is negatively correlated with binocular field of view, meaning that the wider the binocular field of view (better stereoscopic vision), the lower its corrected impact probability (smaller risk).

[0092] For example, correction can be performed using the following formula: P_corrected = P_base × tanh(Speed ​​ / V0) × cos(Angle / 2);

[0093] Where: P_base is the base impact probability before correction (e.g., the single glass block impact probability or building impact probability calculated according to weight 3 or weight 4); Speed ​​is the flight speed of a specific bird, in units consistent with V0 (e.g., meters per second); V0 is a preset speed normalization factor, whose physical meaning is to ensure that the typical flight speed falls within the sensitive variation range of the tanh function. Optionally, V0 = 13.5 m / s, this value can be set with reference to the average flight speed range of common birds (5-20 m / s), so that when the speed is in the range of 0-20 m / s, the output of tanh(Speed / 13.5) monotonically increases from 0 to about 0.9; Angle is the binocular field of view angle of a specific bird, in degrees (°); cos(Angle / 2) is the correction factor for the binocular field of view angle. When Angle = 0°, cos(0°) = 1, the correction factor is at its maximum (no reduction); when Angle = 180°, cos(90°) = 0, the correction factor is 0 (risk is completely eliminated). Since the binocular field of view of actual birds is usually between 10° and 70°, the output range of cos(Angle / 2) is approximately cos(35°) = 0.82 to cos(5°) = 0.996. This factor makes a certain downward correction to the collision probability, but does not completely eliminate the risk. Therefore, even birds with good binocular vision may still collide due to other factors (such as distraction or poor lighting conditions).

[0094] In this embodiment, a risk multiplier with saturation growth is introduced for flight speed using the tanh function. The faster the speed, the shorter the time available for birds to react, and the higher the risk, but there is an upper limit. For binocular field of view, a risk reduction factor related to the angle is introduced using the cosine function. The wider the field of view, the better the stereo vision, and the lower the risk.

[0095] Alternatively, other forms of correction functions can be used, such as linear scaling (P_corrected = P_base × (Speed ​​ / V_max)) or piecewise functions, as long as the corrected impact probability is positively correlated with the flight speed and negatively correlated with the binocular field of view. Furthermore, if only the flight speed is obtained but the binocular field of view is not, the field of view correction term can be omitted (i.e., multiplied only by tanh(Speed / 13.5)); conversely, if only the binocular field of view is obtained but the flight speed is not, the speed correction term can be omitted (i.e., multiplied only by cos(Angle / 2)), thus adapting to application scenarios with different levels of data completeness.

[0096] This embodiment achieves species-specific assessment of impact risk. Compared to schemes that only output a single basic impact probability, this embodiment can distinguish the risk differences of different birds to the same building, providing richer decision-making information for the design of refined impact protection measures (e.g., targeted protection for high-risk species).

[0097] In some embodiments, the number of target regions is multiple; acquiring the original image of the target region includes:

[0098] Glass block detection is performed on the glass curtain wall of the target building, and each detected glass block area is identified as a target area.

[0099] The exterior facade of the target building is surveyed, and original images of each target area are collected.

[0100] After determining the probability of the target flight object colliding with the target area based on at least one of the following: the deployment status of the collision avoidance facilities, the detection results of the flight trajectory associated objects, and the depth information, the method further includes:

[0101] Based on the deployment of the anti-collision facilities, determine the area where no anti-collision facilities are deployed for the target building from among the multiple target areas;

[0102] The total probability of collision between the target building and the target to be detected is determined based on the average value of the collision probability corresponding to the area where no collision avoidance facilities are deployed and the proportion of the area where no collision avoidance facilities are deployed to all the target areas.

[0103] The target building refers to a building that requires impact risk assessment, and its facade may be entirely or partially composed of glass curtain walls. Glass block detection refers to the process of automatically identifying and locating each individual glass panel from acquired images of the building facade using computer vision technology.

[0104] For example, glass block detection can be achieved using deep learning object detection models (such as YOLOv11, Faster R-CNN, etc.). Specifically, a training dataset containing a large number of glass curtain wall images can be pre-constructed, with each glass panel in each image labeled with a bounding box. Then, an object detection model is trained using this dataset. During the actual inference phase, the acquired building facade images are input into the detection model, and the model outputs the positional information of all glass blocks in the image (e.g., the coordinates of the top-left and bottom-right corners of each glass block's bounding box). Each detected glass block region is considered an independent target region.

[0105] Considering the significant differences in impact risk among different glass panels on the same building due to variations in orientation, height, and surrounding environment, it's important to understand that south-facing glass panels directly opposite street trees may pose a high risk, while north-facing panels directly opposite concrete walls may have a low risk. By evaluating each glass panel as an independent target area, this method can generate a detailed risk distribution heatmap of the building's curtain wall, guiding targeted modifications (deploying anti-collision devices only on high-risk glass panels). This significantly reduces modification costs compared to uniformly modifying the entire building. For example, in a curtain wall building with 500 glass panels, if only 50 panels (10%) are in a high-risk area, then these 50 panels can be identified and anti-collision stickers can be deployed only on them.

[0106] Facade patrol refers to the automatic or manual data collection along the perimeter of a target building using a mobile platform. Optionally, a drone can be used as the image acquisition platform. Specifically, the facade patrol route of the target building is planned in advance, and parameters such as the drone's flight altitude, flight speed, distance from the curtain wall, and camera pitch angle are set. For example, the drone's flight altitude can be set to gradually increase from the bottom to the top of the building, the camera lens axis can be set to maintain a small angle with the curtain wall surface to reduce specular reflection interference, and the shooting spacing can be set to ensure sufficient overlap between adjacent images. During flight, the drone automatically triggers the camera to capture raw images of each target area. The acquired raw images are transmitted in real-time or offline to the computing device executing this method. Optionally, other acquisition methods can also be used, such as ground-mounted telephoto cameras, climbing robots, or handheld cameras.

[0107] Areas without deployed collision protection facilities refer to those target areas identified as "without collision protection facilities" during collision protection facility inspection. Specifically, for each glass block area, based on the collision protection facility deployment status label ("deployed" or "not deployed"), all glass block areas labeled "not deployed" are grouped into the set of areas without deployed collision protection facilities. This set is used for subsequent impact probability aggregation calculations. Considering that the risk of areas with deployed collision protection facilities has been reduced to a negligible level, deployed areas should not be included in the aggregation statistics when calculating the overall building risk; otherwise, the risk contribution of high-risk areas will be diluted, leading to a distortion of the overall assessment results. By only counting areas without deployed collision protection facilities, this embodiment can accurately reflect the current "net risk" of the building.

[0108] The average impact probability for areas without crash barriers is calculated by taking the arithmetic mean of the impact probabilities of each of these areas. For example, suppose a building has 100 windows, 30 of which have no crash barriers. The impact probabilities of these 30 windows are 0.85, 0.72, 0.60, ..., 0.15, respectively. Then the average value P_avg = (0.85 + 0.72 + ... + 0.15) / 30.

[0109] The proportion of areas without crash barriers to the total target area refers to the ratio of the number of glass blocks without crash barriers to the total number of glass blocks. Continuing the example, the proportion of areas without crash barriers, R = 30 / 100 = 0.3 (i.e., 30%). Considering that the overall impact risk of a building depends not only on the average risk level of unprotected areas but also on the proportion of unprotected areas in the whole, even if a building has a high average risk in its unprotected areas (e.g., P_avg = 0.9), if the unprotected areas only account for 5% of the total area (R = 0.05), its overall risk may still be lower than another building with a moderate average risk (P_avg = 0.6) but where unprotected areas account for 80% of the total area (R = 0.8). Therefore, it is necessary to integrate both factors to aggregate the risk at the individual glass block level into a single risk indicator at the building level, providing an intuitive and comparable overall risk assessment result.

[0110] Specifically, the formula for calculating the total probability of a building impact is: P_building = P_avg × R. Where P_avg is the average impact probability of areas without deployed collision protection facilities, and R is the proportion of areas without deployed collision protection facilities to all target areas.

[0111] Alternatively, a weighted average can be used: P_building = (Σ P_i) / N_total, where P_i is the impact probability of each glass block (P_i=0 for glass blocks with deployed anti-collision devices), and N_total is the total number of glass blocks. This formula is mathematically equivalent to P_avg × R, because Σ P_i = P_avg × (number of undeployed areas), which, when divided by N_total, yields P_avg × R.

[0112] This embodiment achieves quantitative aggregation from the risk of a single glass block to the risk of the entire building. The total building impact probability can serve as a unified decision-making indicator for comparing risks between different buildings, prioritizing renovations, or as a quantitative criterion for determining compliance with impact-resistant design codes.

[0113] In some embodiments, after determining the probability of the detected flight target colliding with the target area based on at least one of the collision avoidance facility deployment, the detection result of the flight trajectory associated object, and the depth information, the method further includes:

[0114] The collision probability corresponding to the target area is displayed through a preset interactive interface;

[0115] In response to an interactive operation on the impact probability, a collision avoidance facility deployment status is determined based on the interactive operation and the impact probability, representing a collision avoidance modification strategy for a target area where the collision avoidance facility has not been deployed.

[0116] The pre-defined interactive interface refers to a graphical user interface (GUI). This interface receives user input and displays data processing results, presenting the quantitative risk assessment results between the flight target and the target area in a visual and interactive format to relevant parties (such as building owners, property managers, urban planning departments, or ecological protection organizations). This allows the assessment results to be intuitively understood and effectively utilized. For example, the interactive interface can be developed using the Qt framework. Qt is a cross-platform C++ graphical user interface application development framework with excellent cross-platform characteristics and a rich control library, suitable for developing desktop or embedded assessment systems. Alternatively, the interactive interface can also be developed using web front-end technologies (such as HTML5, JavaScript, Vue.js, etc.) and deployed on a cloud server, accessible to users through a browser. This approach is more suitable for multi-user, remote access scenarios.

[0117] The display of impact probability can be diversified to adapt to different users' habits and decision-making needs. For example, the interactive interface can use an overlay display of architectural floor plans or elevations. Specifically, the system first loads a 2D or 3D model of the target building, and then colors and marks each glass block area (target area) on the model: high-risk areas are marked in red, medium-risk areas in yellow, low-risk areas in green, and areas with deployed anti-collision facilities in gray or blue. Users can click on any glass block, and a detailed information window will pop up, displaying the specific impact probability value (e.g., 0.85), semantic recognition results (e.g., "simultaneous presence of sky and trees"), average depth value (e.g., 15.2 meters), deployment status of anti-collision facilities (e.g., "not deployed"), and recommended modification measures (e.g., "recommendation of dotted anti-collision stickers"). This visual display allows users to clearly identify the spatial distribution of high-risk areas at a glance, thereby quickly locating the areas requiring priority intervention.

[0118] In this embodiment, the interactive interface is not merely a passive information display tool, but also a bridge between the user and the evaluation system. By intuitively displaying the collision probability, users can quickly understand the evaluation results and make further decisions based on them. Simultaneously, the interactive interface also provides users with an entry point to input their modification intentions, thereby triggering subsequent strategy generation steps.

[0119] Interactive operations refer to users selecting, filtering, or inputting commands on the displayed impact probabilities through a pre-defined interactive interface. For example, interactive operations may include: clicking the "Generate Modification Plan" button on the interface, selecting a risk level threshold (e.g., "Generate a plan only for high-risk areas"), manually selecting the glass block area to be treated, and entering the upper limit of the modification budget. Collision-avoidance modification strategies refer to specific action plans developed for target areas without deployed collision-avoidance facilities to reduce impact risk. For example, modification strategies may include: the specific locations of glass blocks requiring collision-avoidance facilities, recommended types of collision-avoidance facilities (dot stickers, stripe stickers, UV reflective films, etc.), facility specifications (e.g., dot stickers should have a diameter of 5-10 mm and a spacing of no more than 10 cm), construction priority ranking, and estimated material and labor costs. This combines the user's decision-making intent with the system's quantitative assessment results to automatically generate targeted collision-avoidance modification plans, thereby reducing the difficulty and cost of manual decision-making.

[0120] Considering that the ultimate goal of impact risk assessment is not merely to output a numerical value, but to guide actual collision avoidance modifications, specifically, based on the displayed impact probability distribution and the constraints input by the user through interactive operations (such as budget limitations, construction period, and the area to be protected), a modification strategy is automatically generated.

[0121] For example, the process of determining a collision avoidance retrofit strategy may include: First, obtaining the impact probability P_i of each of the areas without deployed collision avoidance facilities. Then, sorting these areas from highest to lowest impact probability to form a risk priority queue. Second, responding to user interaction, obtaining the user-defined retrofit constraints. For example, users can set the following parameters through the interface: a maximum retrofit budget (e.g., 5000 yuan), the average retrofit cost per glass pane (e.g., 100 yuan per pane for dot-matrix sticker material and labor), the desired minimum risk reduction (e.g., reducing the average impact probability of the retrofitted building to below 0.2), or specifying a maximum number of glass panes to be retrofitted (e.g., no more than 20 panes). Based on the above constraints, high-risk areas are selected sequentially from the risk priority queue, and their retrofit costs are accumulated until the budget limit or other constraints are met. The selected areas are the target areas for this retrofit. The system automatically generates a renovation strategy report, which includes: a list of target renovation areas (with the specific location of each piece of glass and its current impact probability), recommended types and specifications of anti-collision facilities, suggested construction sequence (in order of risk from high to low), estimated total cost, and the expected total building impact probability after renovation (obtained by recalculating P_building after setting P_i of the renovated areas to 0).

[0122] For example, suppose a building has 100 windows, 50 of which lack impact protection. Their impact probabilities range from 0.1 to 0.9. A user sets a renovation budget of 2000 yuan and a renovation cost of 100 yuan per window. The system automatically selects the 20 windows with the highest impact probabilities (20 windows × 100 yuan = 2000 yuan) to generate a renovation strategy. The renovation strategy report will show that the current impact probability of these 20 windows ranges from 0.65 to 0.90, and the expected total building impact probability after renovation will decrease from the current 0.35 to 0.18 (calculated as follows: after renovation, only 30 windows without impact protection remain, with an average impact probability of approximately 0.3, multiplied by the area ratio of 0.3 to get 0.09). Optionally, if the user's budget is insufficient to cover all high-risk areas, the system can also provide "phased renovation suggestions," such as renovating the 10 highest-risk windows in the first phase (budget 1000 yuan), and renovating the remaining medium-to-high-risk areas in the second phase.

[0123] Optionally, the collision avoidance modification strategy can further consider the type of collision avoidance facility. For example, for glass blocks that primarily reflect the sky and trees (high semantic risk), high-contrast white or black dotted stickers are recommended because stickers can effectively disrupt specular reflection. For glass blocks that primarily cause impacts due to transmission (e.g., indoor plants), striped stickers or stickers with ultraviolet reflective properties are recommended, as striped stickers are more easily recognized by birds from a transmission perspective. The system can automatically recommend the type of collision avoidance facility based on semantic recognition results (the presence of sky, trees, indoor plants, etc.) and incorporate this recommendation information into the modification strategy report.

[0124] This embodiment automates the entire process from risk assessment to modification decision-making. Users do not need professional ornithological or architectural knowledge; they can obtain scientific, economical, and feasible collision avoidance modification solutions through simple interactive operations, which can significantly reduce the technical threshold and decision-making costs for promoting and implementing collision avoidance measures.

[0125] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0126] Based on the same inventive concept, this application also provides a glass curtain wall bird impact risk assessment device for implementing the above-mentioned glass curtain wall bird impact risk assessment method. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more glass curtain wall bird impact risk assessment device embodiments provided below can be found in the limitations of the glass curtain wall bird impact risk assessment method above, and will not be repeated here.

[0127] In one exemplary embodiment, such as Figure 3 As shown, a bird strike risk assessment device 300 for glass curtain walls is provided, comprising:

[0128] The acquisition module 302 is used to acquire the original image of the target area;

[0129] Detection module 304 is used to detect anti-collision facilities in the original image to obtain the deployment status of anti-collision facilities in the target area;

[0130] The recognition module 306 is used to perform semantic recognition on the original image to obtain the detection result of the flight trajectory associated object of the flight target to be detected in the original image;

[0131] The estimation module 308 is used to perform depth estimation on the original image to obtain the depth information of the original image;

[0132] The determination module 310 is used to determine the probability of the target object to collide with the target area based on at least one of the following: the deployment status of the collision avoidance facilities, the detection results of the flight trajectory associated objects, and the depth information.

[0133] Each module in the aforementioned bird strike risk assessment device for glass curtain walls can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0134] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 4As shown, the computer device includes a processor, memory, input / output interfaces, a communication interface, a display unit, and an input device. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When executed by the processor, the computer program implements a method for assessing the risk of bird strikes to glass curtain walls. The display unit is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0135] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0136] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps included in any of the foregoing embodiments of the glass curtain wall bird impact risk assessment method.

[0137] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps included in any of the aforementioned embodiments of the glass curtain wall bird impact risk assessment method.

[0138] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps included in any of the foregoing embodiments of the glass curtain wall bird impact risk assessment method.

[0139] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0140] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0141] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0142] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for assessing the risk of bird strikes to glass curtain walls, characterized in that, The method includes: The glass curtain wall of the target building is inspected for glass blocks, and each detected glass block area is identified as a target area; there are multiple target areas. The exterior facade of the target building is surveyed, and original images of each target area are collected. The original image is subjected to collision avoidance facility detection to obtain the collision avoidance facility deployment status of the target area; the collision avoidance facility deployment status includes the coverage of collision avoidance facilities in the target area; Semantic recognition is performed on the original image to obtain the detection results of the flight trajectory association objects of the target to be detected in the original image; the detection results of the flight trajectory association objects include the correlation degree between the flight trajectory association objects existing in the target area and the flight trajectory of the target to be detected; Depth estimation is performed on the original image to obtain the depth information of the original image; the depth information includes the depth value of each pixel in the target region; The probability of the detected flight target colliding with the target area is determined based on at least one of the following: the deployment status of the collision avoidance facilities, the detection results of the flight trajectory associated objects, and the depth information; wherein, if the coverage indicates that the collision avoidance facilities are deployed in the target area, the collision probability is determined based on the coverage; the collision probability is negatively correlated with the coverage of the collision avoidance facilities. If the coverage indicates that no anti-collision facilities are deployed in the target area, the depth value of each pixel is normalized, and the average value of the normalized depth value of all pixels in the target area is calculated. The collision probability is obtained by weighting the average value of the normalized depth values ​​based on the flight trajectory correlation. The collision probability is positively correlated with the flight trajectory correlation between the flight trajectory associated object and the flight trajectory of the target to be detected. The collision probability is also positively correlated with the average depth value of the pixels in the target region. Based on the deployment of the anti-collision facilities, determine the area of ​​the target building that has not been deployed with anti-collision facilities from among the multiple target areas; The total probability of collision between the target building and the target to be detected is determined based on the average value of the collision probability corresponding to the area where no collision avoidance facilities are deployed and the proportion of the area where no collision avoidance facilities are deployed to all the target areas.

2. The method according to claim 1, characterized in that, The flight trajectory association object includes multiple optional association objects; determining the collision probability of the target flight target with the target area based on at least one of the following: the deployment status of the collision avoidance facility, the detection results of the flight trajectory association objects, and the depth information, includes: Determine the co-occurrence of the multiple optional associated objects within the target area; the co-occurrence includes the flight trajectory correlation degree corresponding to each co-occurring optional associated object. The collision probability is determined based on the co-occurrence situation; wherein the collision probability is positively correlated with the maximum value or sum of the correlation degree of the flight trajectories of the co-occurring optional associated objects.

3. The method according to claim 1, characterized in that, The target flight to be detected includes birds; after determining the probability of the target flight to be detected impacting the target area based on at least one of the following: the deployment status of the collision avoidance facilities, the detection results of the flight trajectory associated objects, and the depth information, the method further includes: Obtain the flight speed and / or binocular field of view of the target to be detected; The impact probability is corrected based on the flight speed and / or binocular field of view, wherein the corrected impact probability is positively correlated with the flight speed and negatively correlated with the binocular field of view.

4. The method according to claim 1, characterized in that, After determining the probability of the target flight object colliding with the target area based on at least one of the following: the deployment status of the collision avoidance facilities, the detection results of the flight trajectory associated objects, and the depth information, the method further includes: The collision probability corresponding to the target area is displayed through a preset interactive interface; In response to an interactive operation on the impact probability, a collision avoidance facility deployment status is determined based on the interactive operation and the impact probability, representing a collision avoidance modification strategy for a target area where the collision avoidance facility has not been deployed.

5. A bird strike risk assessment device for glass curtain walls, characterized in that, The device includes: The acquisition module is used to detect glass blocks on the glass curtain wall of the target building and determine each detected glass block area as a target area; there are multiple target areas; the module performs a facade patrol of the target building and collects original images of each target area; The detection module is used to detect collision avoidance facilities in the original image to obtain the deployment status of collision avoidance facilities in the target area; the deployment status of collision avoidance facilities includes the coverage of collision avoidance facilities in the target area; The recognition module is used to perform semantic recognition on the original image to obtain the detection results of the flight trajectory associated objects of the target to be detected in the original image; the detection results of the flight trajectory associated objects include the correlation degree between the flight trajectory associated objects existing in the target area and the flight trajectory of the target to be detected. An estimation module is used to perform depth estimation on the original image to obtain depth information of the original image; the depth information includes the depth value of each pixel in the target region; wherein, if the coverage indicates that the anti-collision facility is deployed in the target region, the collision probability is determined based on the coverage; the collision probability is negatively correlated with the coverage of the anti-collision facility; if the coverage indicates that the anti-collision facility is not deployed in the target region, the depth value of each pixel is normalized, and the average value of the normalized depth values ​​of all pixels in the target region is calculated; the average value of the normalized depth values ​​is weighted according to the flight trajectory correlation to obtain the collision probability; the collision probability is positively correlated with the correlation between the flight trajectory associated object and the flight trajectory of the target to be detected; the collision probability is positively correlated with the average depth value of the pixels in the target region; The determination module is used to determine the area of ​​the target building that has not deployed collision avoidance facilities from multiple target areas based on the deployment status of the collision avoidance facilities; and to determine the total collision probability between the target building and the target to be detected based on the average value of the collision probability corresponding to the area of ​​the target building that has not deployed collision avoidance facilities and the proportion of the area of ​​the target building that has not deployed collision avoidance facilities to all the target areas.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 4.

Images