Strong pet dog safety monitoring system and monitoring method

Abstract

The invention discloses a safety monitoring system and a monitoring method for a strong pet dog, relates to the technical field of dog identification, and solves the problem that field images and infrared imaging cannot be effectively combined to fully improve the identification accuracy. According to the invention, the acquisition modes are intelligently switched according to different illumination conditions, and a clear and high-identification-degree image can be acquired in any environment regardless of a high-definition camera direct acquisition mode in the daytime or an infrared thermal imager imaging mode at night or an innovative combination mode in an intermediate illumination state. The adaptability of image acquisition is greatly improved, and a solid foundation is laid for subsequent accurate identification.

Description

Technical Field

[0001] The present invention relates to the technical field of dog identification, and in particular to a safety monitoring system and a monitoring method for fierce pet dogs. Background Art

[0002] With the acceleration of urbanization, the number of pets kept continues to rise. Among them, the management of fierce pet dogs has become an important part of maintaining urban safety and public order. Accurately monitoring the presence and status of fierce pet dogs is of great significance for preventing potential injuries and protecting the personal safety of residents.

[0003] At present, common pet dog monitoring methods mostly rely on image recognition technology. However, in actual application scenarios, insufficient light environments frequently occur, such as at night, in dark corners indoors, on cloudy days, and other environments, and traditional image recognition technology faces severe challenges; when the light is dim, the images collected by ordinary cameras are often blurred, and the key features of dogs, such as fur texture, body contours, facial features and other details are difficult to present clearly, resulting in the inability to accurately identify dogs, and it is even more difficult to accurately determine whether they are fierce dog breeds.

[0004] Although infrared imaging technology can cope with low-light environments to a certain extent, it is easy to make misjudgments when relying solely on infrared thermal imaging to identify dog ​​features. It is difficult to accurately distinguish different dog breeds and accurately capture the subtle features of dogs. In cases where the light intensity is low, it is not possible to effectively combine on-site images and infrared imaging to fully improve its recognition accuracy, thereby achieving better safety monitoring effects. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention provides a safety monitoring system and method for aggressive pet dogs, which solves the problem that on-site images and infrared imaging cannot be effectively combined to fully improve their recognition accuracy.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: A method for safely monitoring a fierce pet dog comprises the following steps:

[0007] Step 1: Numerical monitoring of the brightness of the monitored area is performed. Based on the monitored brightness, the image acquisition method of the corresponding monitored area is determined and the image is acquired. The acquired image is recorded as the image to be identified. The specific method is as follows:

[0008] Based on the light sensor set in the corresponding monitoring area, the light brightness associated with the corresponding monitoring area is confirmed, and the confirmed light brightness is calibrated as L i , where i represents different monitoring moments;

[0009] The monitored light brightness L iCompare with the preset values Y1 and Y2, where Y1 is the preset credible illuminance limit value of natural light, Y2 is the preset credible illuminance limit value of infrared light, both Y1 and Y2 are luminance data, and Y2 < Y1;

[0010] When L i ≥ Y1, then through the high-definition camera, confirm the specific area image being monitored, and calibrate the specific area image being monitored as the image to be recognized;

[0011] When L i ≤ Y2, then through the infrared thermal imager, confirm the thermal imaging of the specific area being monitored, and calibrate the thermal imaging of the specific area being monitored as the image to be recognized;

[0012] When Y2 < L i < Y1, first confirm the thermal imaging area associated with the current moment according to the infrared thermal imager, and then based on the high-definition camera, confirm the pixel values associated with different points within the corresponding thermal imaging area at the current moment, combine the pixel values with the infrared light intensity for feature combination, and calibrate the combined feature image as the image to be recognized. The specific method is as follows:

[0013] At the current moment, based on the infrared thermal imager, confirm the infrared radiation area associated within the monitoring area, and record the associated infrared radiation area as the area to be determined;

[0014] Calibrate the pixel values associated with different pixel points within different areas to be determined as X i-k , where i represents different areas to be determined, k represents different pixel points, and group several sets of pixel values X i-k belonging to the same area to be determined and sort them in descending order to confirm the pixel value sorting sequence;

[0015] Starting from the first pixel value in the pixel value sorting sequence, select pixel values in turn backward to confirm the pixel value segments, confirm the value difference between adjacent pixel values, and the value difference = |previous set of pixel values - next set of pixel values|. If the value difference ≤ X1, then calibrate the adjacent pixel values as the same type of pixel values, and so on, gradually confirm the same type of pixel values backward. When the confirmed value difference satisfies: value difference > X1, stop calibrating the same type of pixel values, and calibrate the confirmed multiple sets of the same type of pixel values as the same value segment, and then record the confirmed same value segment as the pixel value segment, where X1 is the preset pixel fluctuation value, and record the pixel positions associated with the current pixel value segment as the pixel feature positions;

[0016] Then confirm the infrared light intensity associated with other points within this area to be determined, execute several feature combination processes, and confirm the combined features associated with different feature combination processes, and confirm the best process based on the combined features. The specific method is as follows:

[0017] Process 1: Calibrate several groups of pixel feature points associated with the pixel value segment, perform mean processing on several groups of pixel values associated with the several groups of pixel feature points to confirm the pixel mean feature T1. Then, perform mean processing on the infrared light intensities associated with other points confirmed within the undetermined area to confirm the processed mean T2. Use: JT q = T1 × C1 + T2 × C2 to confirm the combined feature JT associated with this combined process q , where q represents different combined processes, and both C1 and C2 are preset fixed coefficient factors;

[0018] Process 2: Remove the last group of pixel feature points from the pixel value segment. The last group of pixel feature points will be classified into other points within the undetermined area. Use the same processing method as in Process 1 to process and confirm the combined feature JT associated with this process q ;

[0019] And so on. Each time a group of processes is executed, the last group of pixel feature points is removed from the pixel value segment until, in the last group of processes, there are only two groups of pixel feature points left inside the pixel value segment. Select from the combined features associated with several groups of feature combination processes and select the feature combination process associated with JT q max as the best process;

[0020] Display the pixel feature points and other points associated with the best process using different images. The area corresponding to the pixel feature points is displayed in the form of a picture, and the area corresponding to other points is displayed in the form of an infrared thermal image. Combine the two displayed pictures to confirm the combined feature image;

[0021] Step 2: Based on a preset fierce pet dog recognition model, confirm whether there is a fierce dog in the image to be recognized. If there is, generate a fierce dog signal; if not, do not generate any signal. The specific method is as follows:

[0022] The fierce pet dog recognition model includes an image analysis module and an infrared light analysis module. The image analysis module performs fierce dog recognition on the image to be recognized in the form of a picture and outputs a fierce dog index. If the fierce dog index ≥ 0.8, it means there is a fierce dog and a fierce dog signal is output; otherwise, no signal is output;

[0023] The infrared light analysis module performs fierce dog recognition on the image to be recognized in the form of a thermal image and outputs a fierce dog index. If the fierce dog index ≥ 0.8, it means there is a fierce dog and a fierce dog signal is output; otherwise, no signal is output;

[0024] If there are two groups of image regions in the image to be recognized, namely the picture form and the thermal imaging form, confirm the proportion of the two different forms of image regions in the entire image to be recognized. Use: ZB1 = area of the picture form image ÷ total area of the image to be recognized, and ZB2 = area of the thermal imaging form image ÷ total area of the image to be recognized. Use the image analysis module to identify the vicious dogs in the picture form image region and output the vicious dog index K1. Then use the infrared light analysis module to identify the vicious dogs in the thermal imaging form image region and output the vicious dog index K2. Use Kz = K1×ZB1 + K2×ZB2 to confirm the total coefficient. If Kz≥0.8, it means there are vicious dogs, then output the vicious dog signal; otherwise, do not output.

[0025] Step 3: According to the vicious dogs detected in the image to be recognized, confirm the surrounding images of the vicious dogs, and identify whether there are leash features in the surrounding images. If there are leash features, keep the original vicious dog signal unchanged. If there are no leash features, generate a vicious dog danger signal for display. The specific method is as follows:

[0026] S31: Based on the vicious dogs detected in the image to be recognized, use this vicious dog as the reference feature, confirm the edge contour of the reference feature, and use this edge contour and the center point of this edge contour as the reference to expand the range and confirm a group of expanded regions. The width of the expanded range of the expanded region is twice the original edge contour.

[0027] S32: Use the Sobel algorithm to confirm the gradient pixels at different points in the expanded region, and label the gradient pixels associated with different points as TD m , where m represents different pixel points in the expanded region. Label the pixel points where TD m > X2 as gradient pixel points; otherwise, do not perform any labeling. X2 is a preset value.

[0028] Sequentially confirm the contours associated with adjacent gradient pixel points, label one by one the multiple range contours involved in the expanded region, and record the range contour in contact with the edge contour of the vicious dog as the pending contour.

[0029] S33: Randomly select an endpoint of a contour on the pending contour as the starting point, determine the nearest point on the other contour of the starting point, and record the distance value. Then sequentially confirm the subsequent points from the starting point and synchronously confirm the distance values of the nearest points. And so on. If the sequentially confirmed distance values all belong to the preset interval, and the preset interval is confirmed according to the width of the leash on the current market, then label this pending contour as the leash feature and keep the original vicious dog signal unchanged; otherwise, generate a vicious dog danger signal for display.

[0030] The present invention provides a safety monitoring system and a monitoring method for fierce pet dogs. Compared with the prior art, the following beneficial effects are achieved:

[0031] In the image acquisition stage, the acquisition method is intelligently switched according to different lighting conditions. Whether it is direct acquisition by a high-definition camera during the day, imaging by an infrared thermal imager at night, or an innovative combination method in an intermediate lighting state, clear and highly recognizable images can be obtained in any environment. This greatly improves the adaptability of image acquisition and lays a solid foundation for subsequent accurate recognition;

[0032] For the recognition of fierce dogs, based on a recognition model trained with a large amount of data, covering image analysis and infrared light analysis modules, it can not only accurately recognize fierce dogs in different forms of images, but also, when faced with complex images (simultaneously including a video image and a thermal imaging area), obtain a comprehensive judgment result through scientific weighted calculation, greatly improving the accuracy and reliability of recognition and effectively reducing misjudgment and missed judgment;

[0033] By expanding the range based on the edge contour of the fierce dog, using the Sobel algorithm to screen gradient pixel points to determine the contour, and then judging whether it is a leash feature according to a preset interval, it is possible to efficiently and accurately judge whether the fierce dog is in a safe control state. Once the situation of no leash is recognized, a danger signal is sent in a timely manner, effectively ensuring the safety of public areas, reducing the risk of fierce dogs hurting people, and maintaining social order and the safety of the people. Overall, the monitoring method is comprehensive and detailed, providing a complete and efficient solution for the safety management of fierce pet dogs from image acquisition to recognition and then to the determination of the safe state. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0035] Figure 2 It is a schematic diagram of the principle of the fierce pet dog recognition model of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0036] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0037] Embodiment 1

[0038] Please refer to Figure 1 , this application provides a safety monitoring method for fierce pet dogs, including the following steps:

[0039] Step 1: Numerically monitor the light brightness of the monitored area. Based on the monitored light brightness, confirm the image acquisition method for the corresponding monitored area and acquire the image. Denote the acquired image as the image to be recognized. Specifically, for day and night, the image acquisition methods are different. During the day, the light is strong, and generally, the image acquisition method can be directly used to acquire the area image, and a high-definition camera can be used to acquire the picture. For night, the brightness is low, and generally, only an infrared thermal imager can be used to display the thermal image to confirm the infrared thermal imaging of the corresponding monitored area and identify and confirm such thermal imaging;

[0040] Among them, the specific method for image acquisition is as follows:

[0041] Based on the light sensor set for the corresponding monitored area, confirm the light brightness associated with the corresponding monitored area, and calibrate the confirmed light brightness as L i , where i represents different monitoring times;

[0042] Compare the monitored light brightness L i with the preset values Y1 and Y2. Among them, Y1 is the preset credible illuminance limit value of natural light, and Y2 is the preset credible illuminance limit value of infrared light. The specific values of Y1 and Y2 are determined in advance by the operator. Both Y1 and Y2 are brightness data, and Y2 < Y1;

[0043] When L i ≥ Y1, then through the high-definition camera, confirm the specific area picture of the monitoring, and calibrate the specific area picture of the monitoring as the image to be recognized;

[0044] When L i ≤ Y2, then through the infrared thermal imager, confirm the specific area thermal imaging of the monitoring, and calibrate the specific area thermal imaging of the monitoring as the image to be recognized;

[0045] When Y2 < L i < Y1, first confirm the thermal imaging area associated with the current time according to the infrared thermal imager, and then based on the high-definition camera, confirm the pixel values associated with different points in the corresponding thermal imaging area at the current time. Combine the pixel values with the infrared light intensity, and calibrate the combined feature image as the image to be recognized. Specifically, in order to make the features to be recognized more obvious, the corresponding image and infrared imaging features can be confirmed in a darker state, combine the corresponding image features, and then perform feature recognition on the combined image, which can improve the feature recognition accuracy of the corresponding combined image, achieve a more obvious image recognition effect of the features, and ensure the specific accuracy of the fierce dog;

[0046] When Y2 < L iWhen < Y1, the specific method of combining the pixel value with the infrared light intensity is as follows:

[0047] At the current moment, based on the infrared thermal imager, confirm the associated infrared radiation area within the monitoring area, and record the associated infrared radiation area as the area to be determined;

[0048] Calibrate the pixel values associated with different pixel points in different areas to be determined as X i-k , where i represents different areas to be determined, k represents different pixel points, and group several sets of pixel values X i-k belonging to the same area to be determined. Sort them in descending order to confirm the pixel value sorting sequence, and the numerical range of the pixel values is between 0 and 255;

[0049] Starting from the first pixel value in the pixel value sorting sequence, select pixel values one by one backward to confirm the pixel value segment, and confirm the value difference between adjacent pixel values. The value difference = |the previous set of pixel values - the next set of pixel values|. If the value difference ≤ X1, then mark the adjacent pixel values as the same type of pixel values, and so on. Gradually confirm the same type of pixel values backward. When the confirmed value difference satisfies: the value difference > X1, stop calibrating the same type of pixel values, and mark the confirmed multiple sets of the same type of pixel values as the same value segment, and then record the confirmed same value segment as the pixel value segment, where X1 is the preset pixel fluctuation value, and its specific value is determined by the operator according to experience. Generally, X1 takes the value of 30, and mark the pixel positions associated with the current pixel value segment as pixel feature positions;

[0050] Then, confirm the infrared light intensity associated with other positions within this area to be determined (here, other positions refer to other positions except the pixel feature positions, and other positions and pixel feature positions are all within this area to be determined), and execute several feature combination processes:

[0051] Process 1: Calibrate several groups of pixel feature positions associated with the pixel value segment, and perform mean processing on several groups of pixel values associated with several groups of pixel feature positions to confirm the pixel mean feature T1. Then, perform mean processing on the infrared light intensity associated with other positions confirmed within the area to be determined to confirm the processed mean T2. Use: JT q = T1 × C1 + T2 × C2 to confirm the combination feature JT associated with this combination process q , where q represents different combination processes, and both C1 and C2 are preset fixed coefficient factors, and their specific values are determined by the operator according to experience;

[0052] Process 2: Exclude the last group of pixel feature positions from the pixel value segment, and the last group of pixel feature positions will be classified into other positions within the area to be determined. Use the same processing method as in Process 1 to process and confirm the combination feature JT associated with this processq ;

[0053] And so on. For each set of processes executed, the last set of pixel feature points is removed from the pixel value segment. Until in the last set of processes, when there are only two sets of pixel feature points inside the pixel value segment, the combined features associated with several sets of feature combination processes are selected for the process, and the feature combination process associated with JT q max is denoted as the optimal process;

[0054] The pixel feature points and other points associated in the optimal process are displayed using different images. The area corresponding to the pixel feature points is displayed in the form of a picture, and the area corresponding to other points is displayed in the form of an infrared thermal image. The two displayed pictures are combined to confirm the combined feature image;

[0055] Specifically, different processing methods are used for specific confirmation of the regional pictures in different lighting environments. And in the subsequent picture analysis process, different recognition models are used to recognize different types of pictures to confirm the existence probability of fierce dogs. When the light condition is in the middle of the set threshold, then feature analysis is required to determine the strongest feature performance state, so as to ensure the accuracy of subsequent specific picture recognition;

[0056] The pixel value representations and the related representations of the infrared light intensity at the corresponding points in the corresponding area are combined in different types to ensure that each area in the combination process can obtain a strong feature display, either the pixel value feature or the numerical intensity feature of the infrared light. Whichever processing feature is used, the accuracy can be guaranteed in the subsequent recognition environment.

[0057] Step 2: Based on a preset fierce pet dog recognition model, confirm whether there is a fierce dog in the image to be recognized. If there is, generate a fierce dog signal. If not, do not generate any signal;

[0058] Specifically, before the preset fierce pet dog recognition model performs image recognition, it needs to go through a large number of trainings. Its training steps are as follows:

[0059] Construct and preprocess the data set: Collect image and video data containing fierce pet dogs and non-fierce pet dogs, label the data, and the labeling content includes the bounding box and class label of the dog (such as "fierce dog" or "non-fierce dog"). Preprocess the data, including image scaling, normalization, data augmentation (such as rotation, flipping, cropping, etc.) to improve the generalization ability of the model;

[0060] Building a deep learning model: Use an object detection algorithm (such as YOLO, SSD, or Faster R-CNN) as the base model, and optimize and adjust the model according to the specific requirements of fierce dog recognition. For example, modify the output layer to adapt to the number of categories, and use a pre-trained model (such as ResNet, EfficientNet) for transfer learning to accelerate model training and improve performance;

[0061] Training and optimizing the model: Define a loss function (such as cross-entropy loss function) and an optimizer (such as Adam or SGD), train the model using the labeled dataset, update the model parameters through the backpropagation algorithm, evaluate the model performance on the validation set, and adjust the hyperparameters (such as learning rate, batch size) to optimize the model accuracy and generalization ability;

[0062] Monitoring and recognition process: Deploy the trained model to the monitoring system, analyze the video stream data in real time, perform object detection on each frame of the image, identify the dog and determine whether it is a fierce pet dog. When a fierce pet dog is detected, trigger an alarm or notify the relevant personnel.

[0063] Since the method of analyzing pet categories through data models is relatively common in the prior art, it will not be elaborated here. When performing relevant analysis on infrared imaging, just replace the data required for training the corresponding model with the relevant data of infrared imaging;

[0064] The specific sub-steps for confirming whether there is a fierce dog in the image to be recognized are as follows:

[0065] The fierce pet dog recognition model includes an image analysis module and an infrared light analysis module. The image analysis module recognizes fierce dogs in the image to be recognized in the form of a picture and outputs a fierce dog index. If the fierce dog index ≥ 0.8, it means there is a fierce dog and a fierce dog signal is output. Otherwise, no output is made. The fierce pet dog recognition model is generally trained based on the yolov5 model, and the value output by the model recognition is a recognition probability value. Combined with Figure 2, the specific recognition process of its model is generally as follows: based on the associated monitoring nodes, the transmitted video image data and on-site illuminance parameters are processed, and the canine judgment results are sent to the image analysis module and the infrared light analysis module; both the image analysis module and the infrared light analysis module are trained based on the YOLOv5 model. By analyzing and identifying the dog breed and whether it is a vicious dog in the corresponding band images, the images for which the dog breed cannot be judged are respectively stored in the corresponding undetermined libraries for further optimization. By comprehensively judging the specific classification results of the dog breed, the subsequent risk grading and early warning module outputs the corresponding vicious dog presence signal, conducts specific recognition and analysis on the natural light images and infrared light images, confirms the dog breed type according to the specific recognition results and outputs it. The natural light image is the relevant image in the form of a picture;

[0066] The infrared light analysis module conducts vicious dog recognition on the image to be recognized in the form of thermal imaging and outputs the vicious dog index. If the vicious dog index ≥ 0.8, it means there is a vicious dog, and then the vicious dog signal is output; otherwise, it is not output.

[0067] If there are two groups of image areas in the form of a picture and thermal imaging in the image to be recognized, confirm the proportion of the areas of the two different forms of image areas in the entire image to be recognized. Adopt: ZB1 = the area of the image in the form of a picture ÷ the total area of the image to be recognized, and ZB2 = the area of the image in the form of thermal imaging ÷ the total area of the image to be recognized. The image analysis module conducts vicious dog recognition on the image area in the form of a picture and outputs the vicious dog index K1. Then, the infrared light analysis module conducts vicious dog recognition on the image area in the form of thermal imaging and outputs the vicious dog index K2. Adopt Kz = K1×ZB1 + K2×ZB2 to confirm the total coefficient. If Kz ≥ 0.8, it means there is a vicious dog, and then the vicious dog signal is output; otherwise, it is not output.

[0068] Adopting this kind of recognition and analysis processing method can not only display the results of the monitoring images in the form of a picture, but also display the results of the monitoring images in the form of thermal imaging at the same time. It can also conduct numerical analysis on the combined images in the form of a picture and thermal imaging, confirm the result signal, and conduct result display, which can effectively ensure the specific accuracy of its recognition.

[0069] Step 3: This step is only for the image to be recognized in the form of a picture. According to the vicious dog detected in the image to be recognized, confirm the surrounding images of the vicious dog, and identify whether there is a leash feature in the surrounding images. If there is a leash feature, keep the original vicious dog signal unchanged. If there is no leash feature, generate a vicious dog danger signal for display. Among them, the specific method for identifying whether there is a leash feature is:

[0070] S31. Based on the monitored vicious dog in the image to be recognized, use this vicious dog as a reference feature, confirm the edge contour of the reference feature, and use this edge contour and the center point of this edge contour as a reference to expand the range, confirm a set of expanded areas. The center point of its edge contour can be confirmed based on a two-dimensional coordinate system. Combine the two-dimensional coordinate system with the edge contour to confirm the two-dimensional coordinates associated with different contour points within the edge contour. Then perform mean processing on several sets of two-dimensional coordinates to confirm the mean coordinate. The mean coordinate is the position where the center point of the edge contour is located. The width of the expanded range of the expanded area is twice that of the original edge contour. For example, assume the straight-line distance from the center point to a point on the edge contour is L1. After the range is expanded, the associated straight-line distance of this point is 2L1;

[0071] S32. Use the Sobel algorithm to confirm the gradient pixels at different points within the expanded area, and label the gradient pixels associated with different points as TD m , where m represents different pixel points within the expanded area. Label the pixel points where TD m > X2 as gradient pixel points. Otherwise, do not perform any labeling. X2 is a preset value, and its specific value is determined by the operator according to experience;

[0072] Sequentially confirm the contours associated with adjacent gradient pixel points, label one by one the multiple range contours involved within the expanded area, and mark the range contour in contact with the edge contour of the vicious dog as the pending contour;

[0073] Randomly select an endpoint of a contour on the pending contour as the starting point, determine the nearest point on the other side of the contour, and record the distance value. Then, sequentially confirm the subsequent points starting from the starting point and simultaneously confirm the distance values of the nearest points. And so on. If the sequentially confirmed distance values all fall within a preset interval, the preset interval is determined by the operator according to experience and is confirmed based on the width of the leash currently on the market, then mark this pending contour as the leash feature, and keep the original vicious dog signal unchanged. Otherwise, generate a vicious dog danger signal for display.

[0074] According to the edge contour of the corresponding vicious dog, expand and confirm the range, and then perform feature recognition within the expanded range to confirm whether there is a corresponding leash feature. Based on the sequential confirmation of gradient pixel points, the relevant contours existing within the corresponding area can be quickly confirmed. Then, based on the specific confirmation of the width value, to evaluate whether such contours belong to the relevant features of the leash, for the safety monitoring and early warning of vicious dogs.

[0075] Vicious pet dog safety monitoring system, including:

[0076] The image processing end to be recognized monitors the brightness value of the monitored area. Based on the monitored brightness, it confirms the image acquisition method for the corresponding monitored area and acquires the image, and records the acquired image as the image to be recognized;

[0077] The fierce dog model recognition end confirms whether there is a fierce dog in the image to be recognized based on the preset fierce pet dog recognition model. If there is, it generates a fierce dog signal; if not, it does not generate any signal;

[0078] The leash feature recognition end confirms the surrounding images of the fierce dog according to the fierce dog detected in the image to be recognized, and recognizes whether there is a leash feature in the surrounding images. If there is a leash feature, it keeps the original fierce dog signal unchanged; if there is no leash feature, it generates a fierce dog danger signal for display.

[0079] Some of the data in the above formula are numerically calculated after removing their dimensions, and the content not described in detail in this specification belongs to the prior art well-known to those skilled in the art.

[0080] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A method for safely monitoring aggressive pet dogs, characterized in that: The following steps are involved: Step 1: numerically monitor the light brightness of the monitoring area, determine the image acquisition method of the corresponding monitoring area based on the monitored light brightness, and acquire the image, and record the acquired image as the image to be identified; Step 2: Based on the preset fierce pet dog recognition model, confirm whether there is a fierce dog in the image to be recognized. If so, generate a fierce dog signal; if not, do not generate any signal; Step 3: Based on the fierce dog detected in the image to be identified, confirm the surrounding image of the fierce dog, and identify whether there is a leash feature from the surrounding image. If there is a leash feature, keep the original fierce dog signal unchanged; if there is no leash feature, generate a fierce dog danger signal for display.

2. The method for safely monitoring aggressive pet dogs according to claim 1, characterized in that: In step 1, the specific method of confirming the image acquisition method of the corresponding monitoring area and performing image acquisition is: Based on the light sensor set in the corresponding monitoring area, the light brightness associated with the corresponding monitoring area is confirmed, and the confirmed light brightness is calibrated as L i , where i represents different monitoring moments; The monitored light brightness L i Compare with preset values ​​Y1 and Y2, where Y1 is the preset credible illuminance limit value of natural light, and Y2 is the preset credible illuminance limit value of infrared light, where Y1 and Y2 are both brightness data, and Y2<Y1; When L i When ≥Y1, the specific area being monitored is confirmed through a high-definition camera, and the specific area being monitored is marked as the image to be identified; When L i When ≤Y2, the thermal image of the specific monitored area is confirmed by the infrared thermal imager, and the thermal image of the specific monitored area is calibrated as the image to be identified; When Y2<L i When <Y1, the thermal imaging area associated with the current moment is confirmed based on the infrared thermal imager first, and then the pixel values ​​associated with different points in the corresponding thermal imaging area at the current moment are confirmed based on the high-definition camera, and the pixel values ​​and infrared light intensity are combined to perform feature combination, and the combined feature image is calibrated as the image to be identified.

3. The method for safely monitoring aggressive pet dogs according to claim 2, characterized in that: When Y2<L i When <Y1, the specific method of combining the pixel value and the infrared light intensity is as follows: At the current moment, the infrared radiation area associated with the monitoring area is confirmed based on the infrared thermal imager, and the associated infrared radiation area is recorded as a pending area; The pixel values ​​associated with different pixel points in different undetermined areas are calibrated as X i-k , where i represents different undetermined regions, k represents different pixels, and several groups of pixel values ​​X belonging to the same undetermined region are i-k Sort by large to small to confirm the pixel value sorting sequence; Starting from the first pixel value in the pixel value sorting sequence, pixel values ​​are selected in sequence to confirm the pixel value segment, and the value difference between adjacent pixel values ​​is confirmed, where the value difference = |previous group of pixel values ​​- next group of pixel values|. If the value difference ≤ X1, the adjacent pixel values ​​are calibrated as the same type of pixel values. Similarly, the same type of pixel values ​​are confirmed step by step. When the confirmed value difference satisfies: value difference > X1, the calibration of the same type of pixel values ​​is stopped, and the confirmed multiple groups of the same type of pixel values ​​are calibrated as the same value segment, and then the confirmed same value segment is recorded as the pixel value segment, where X1 is the preset pixel fluctuation value, and the pixel point associated with the current pixel value segment is recorded as the pixel feature point; Then, the infrared light intensity associated with other points in the undetermined area is confirmed, several feature combination processes are executed, and the combined features associated with different feature combination processes are confirmed, and the best process is confirmed based on the combined features; The pixel feature points and other points associated with the optimal process are displayed using different images. The areas corresponding to the pixel feature points are displayed in the form of pictures, and the areas corresponding to other points are displayed in the form of infrared thermal imaging. The two sets of displayed pictures are combined to confirm the combined feature image.

4. The method for safely monitoring aggressive pet dogs according to claim 3, characterized in that: The specific method of determining the best process based on the combined features is: Process 1: calibrate several groups of pixel feature points associated with the pixel value segment, and average several groups of pixel values ​​associated with several groups of pixel feature points to confirm the pixel average feature T1, and then confirm the infrared light intensity associated with other points in the undetermined area and average it to confirm the processing average T2, using: JT q =T1×C1+T2×C2 confirms the combined feature JT associated with this combined process q , where q represents different combination processes, and C1 and C2 are both preset fixed coefficient factors; Process 2: Eliminate the last set of pixel feature points from the pixel value segment. The last set of pixel feature points is divided into other points in the pending area and processed in the same way as process 1 to confirm the combined feature JT associated with this process. q ; Similarly, each time a set of processes is executed, the last set of pixel feature points is removed from the pixel value segment. When there are only two sets of pixel feature points in the pixel value segment of the last set of processes, the combined features associated with several sets of feature combination processes are selected for the process, and JT is selected. q The feature combination process associated with max is recorded as the best process.

5. The method for safely monitoring aggressive pet dogs according to claim 1, characterized in that: In step 2, the specific method of identifying a fierce dog is as follows: The fierce pet dog identification model includes an image analysis module and an infrared light analysis module. The image analysis module identifies fierce dogs on the image to be identified in the form of a screen and outputs a fierce dog index. If the fierce dog index is ≥ 0.8, it means that there is a fierce dog and a fierce dog signal is output. Otherwise, no output is made. The infrared light analysis module identifies the fierce dog in the image to be identified in the form of thermal imaging, and outputs the fierce dog index. If the fierce dog index is ≥ 0.8, it means that there is a fierce dog, and the fierce dog signal is output. Otherwise, no output is made. If there are two groups of image areas in the form of picture and thermal imaging simultaneously in the image to be identified, the area ratio of the two image areas in different forms in the entire image to be identified is confirmed by using: ZB1 = picture form image area ÷ total area of ​​the image to be identified and ZB2 = thermal imaging form image area ÷ total area of ​​the image to be identified. The image analysis module is used to identify the fierce dog in the image area in the picture form, and the fierce dog index K1 is output. The infrared light analysis module is used to identify the fierce dog in the image area in the thermal imaging form, and the fierce dog index K2 is output. Kz = K1 × ZB1 + K2 × ZB2 is used to confirm the total coefficient. If Kz ≥ 0.8, it means that a fierce dog exists, and a fierce dog signal is output. Otherwise, no output is made.

6. The method for safely monitoring aggressive pet dogs according to claim 1, characterized in that: The step three is only for the image to be recognized in the form of a picture.

7. The method for safely monitoring aggressive pet dogs according to claim 1, characterized in that: In step 3, the specific method of identifying whether there is a traction rope feature is: S31, based on the fierce dog detected in the image to be identified, the fierce dog is used as a reference feature, and the edge contour of the reference feature is confirmed, and the edge contour and the center point of the edge contour are used as a reference to expand the range and confirm a set of expanded areas, and the expanded range width of the expanded area is twice the original edge contour; S32, using the Sobel algorithm to confirm the gradient pixels at different points in the expanded area, and calibrate the gradient pixels associated with different points as TD m , where m represents different pixels in the expanded area, and TD m >X2 pixels are calibrated as gradient pixels, otherwise, no calibration is performed and X2 is the preset value; Confirm the contours associated with adjacent gradient pixel points in sequence, calibrate the multiple range contours involved in the expanded area one by one, and record the range contours that are in contact with the edge contour of the fierce dog as pending contours; S33. Randomly select an endpoint of a contour from the pending contour and record it as the starting point, determine the nearest point of the starting point on the contour on the other side, and record the distance value, then confirm the subsequent points in sequence from the starting point and simultaneously confirm the distance value of the nearest point, and so on. If the distance values ​​confirmed in sequence all belong to the preset interval, and the preset interval is confirmed according to the width of the traction rope currently on the market, then this pending contour is recorded as the traction rope feature, and the original fierce dog signal is kept unchanged. Otherwise, a fierce dog danger signal is generated for display.

8. A system for monitoring the safety of aggressive pet dogs, the system being operated according to the method for monitoring the safety of aggressive pet dogs according to any one of claims 1 to 7, characterized in that: include: The image processing end to be identified monitors the brightness of the monitored area, determines the image acquisition method of the corresponding monitored area based on the monitored brightness, and acquires the image, and records the acquired image as the image to be identified; The fierce dog model recognition end confirms whether there is a fierce dog in the image to be recognized based on the preset fierce pet dog recognition model. If so, a fierce dog signal is generated; if not, no signal is generated; The leash feature recognition end confirms the surrounding images of the fierce dog based on the fierce dog monitored in the image to be identified, and identifies whether there is a leash feature from the surrounding images. If the leash feature exists, the original fierce dog signal remains unchanged. If the leash feature does not exist, a fierce dog danger signal is generated for display.

Images