Intelligent watering control method for road environmental sanitation

Abstract

The invention relates to the technical field of program control, in particular to an intelligent watering control method for road environmental sanitation. The method comprises the following steps: acquiring an image sequence of pedestrians on a road on a predetermined watering path through a camera; according to the image sequence of the road pedestrians, whether the road pedestrians pull the pet or not is judged through a preset target recognition model; if a road pedestrian pulls the pet, the road pedestrian pulling the pet is marked as a pulling person, and when the pulling person is approached, a preset front watering control action is started to be executed in advance, so that the interference of the pulled pet on the pulling person is reduced. According to the intelligent watering control method for road environmental sanitation, pedestrians beside the road can be better avoided.

Description

Technical Field

[0001] The present application relates to the field of program control technology, and in particular to a method for controlling intelligent watering of road sanitation. Background Art

[0002] At present, in order to ensure the safety of pedestrians, road sanitation sprinkler trucks are usually equipped with cameras or sensors to detect pedestrians on the road ahead when performing watering operations. When a pedestrian is detected, the sprinkler truck will automatically stop watering to avoid inconvenience or safety hazards to pedestrians. However, existing sprinkler control methods are mainly based on the recognition of pedestrian positions, assuming that pedestrian behavior is predictable and controllable.

[0003] But in real life, there are pedestrians who use leashes to lead their pet dogs on the road. The behavior of pet dogs is uncertain and easily stimulated by the external environment. When a sprinkler truck approaches, the alarm, sprinkler sound or water flow may scare the pet dog, causing it to suddenly change direction or speed. In this case, if the tractor is avoided according to the normal range of pedestrians stopping sprinkler, due to the small range of stopping sprinkler, the pet dog may pull the pedestrian with the leash, causing the pedestrian to accidentally enter the sprinkler area, or the pedestrian may lose balance in the process of controlling the pet and avoiding entering the sprinkler area, causing non-contact accidents such as falls.

[0004] Therefore, the prior art still has some problems, which makes it impossible to better avoid pedestrians during watering operations. Summary of the invention

[0005] In order to solve the above technical problems or at least partially solve the above technical problems, the present application provides a road sanitation intelligent sprinkler control method, which can better avoid pedestrians on the road.

[0006] The present application provides a road sanitation intelligent sprinkler control method, comprising the following steps: Acquire an image sequence of pedestrians on a predetermined watering path through a camera; Based on the image sequence of pedestrians on the road, a preset target recognition model is used to judge whether the pedestrians on the road are pulling the pet; If there is a pedestrian on the road pulling his pet, the pedestrian pulling the pet is marked as the puller, and the predetermined pre-sprinkling control action is executed in advance when approaching the puller to reduce the interference of the pulled pet on the puller.

[0007] Optionally, the following steps are also included: Calculate the distance to the tractor to start watering in advance: Obtaining a body size ratio between the towing person and the towed pet, and calculating a pre-risk value according to the body size ratio by using a preset pre-risk value calculation function, wherein the pre-risk value calculation function is set so that the closer the body size of the towed pet is to the body size of the towing person, the higher the pre-risk value; According to the pre-risk value, the distance to the tractor that needs to be controlled for watering in advance is calculated by a preset control watering distance calculation function, and the control watering distance calculation function is set so that the larger the value of the pre-risk value is, the larger the distance that needs to be controlled for watering in advance is; The predetermined pre-watering control action is started at the distance ahead of the towing person at which the watering control needs to be started in advance.

[0008] Optionally, the body size ratio of the pulling person and the pulled pet is calculated by the following steps: When the pulling person is detected, an image sequence of the pulling person in a first time window is acquired; For each photo in the first time window, extract the outlines of the person pulling the pet; Calculate the area of ​​the outline of the towing person and the towed pet in each photo, and obtain the body size ratio of the towing person and the towed pet by using a preset body size ratio calculation function; The preset body shape ratio calculation function is: Where I is the total number of photos in the first time window, i is the i-th photo in the first time window, A(p,i) is the outline area of ​​the pulled pet in the i-th photo, and A(m,i) is the outline area of ​​the pulling person in the i-th photo; in, It is the average of the contour areas of the pulled pet in all photos within the entire first time window.

[0009] Optionally, the following steps are also included: The predetermined pre-sprinkler control action includes an interference enhancement phase and an interference reduction phase. When the predetermined pre-sprinkler control action is started, the interference enhancement phase of a predetermined duration is first entered. In the interference enhancement phase, the interference stimulus to the outside during the sprinkling is enhanced. After exiting the interference enhancement phase, the interference reduction phase is entered to reduce the interference stimulus to the outside during the sprinkling until the position of the tractor is reached. According to the visible light image sequence of the tractor when executing the predetermined pre-sprinkling control action and the pre-risk value, the distance to be controlled for sprinkling after the tractor is calculated: In the interference enhancement stage, according to the visible light image sequence of the tractor when executing the predetermined pre-sprinkling control action, the posture change of the tractor is obtained through a preset posture recognition model to determine whether to exit the interference enhancement stage at present; According to the length of the interference enhancement phase, the pre-risk value is changed by a preset post-risk value calculation function to obtain a post-risk value, so that the longer the interference enhancement phase is, the lower the post-risk value obtained is; According to the post-risk value, the distance at which watering needs to be controlled after the person is pulled is calculated by a preset watering control distance calculation function; The interference reduction stage is maintained until the distance after the tractor leaves the tractor and the water sprinkling needs to be controlled, and then the interference reduction stage is exited.

[0010] Optionally, the preset posture recognition model is trained by the following steps: Constructing the posture recognition model framework, the posture recognition model framework includes an input layer, a skeleton point feature extraction layer, a feature integration layer, a convolutional neural network layer and an output layer; An input layer, for receiving the visible light image of the traction person after entering the interference enhancement stage; The skeleton point feature extraction layer is a preset skeleton key point extraction algorithm, which is used to receive the data output by the input layer and calibrate the skeleton key points of the tractor in the visible light image of the tractor, and the skeleton key points include shoulders, elbows and legs; The feature integration layer is used to create a sliding input window of a preset size, which is used to receive the input of the bone point feature extraction layer within a preset size time to form a bone point feature change sequence for external output; The convolutional neural network layer is used to receive the skeleton point feature change sequence transmitted by the feature integration layer, perform data processing on the skeleton point feature change sequence and then output it externally; The output layer is used to obtain the output of the convolutional neural network layer, process the output of the convolutional neural network layer and output it externally; Collect image sequences of multiple people pulling pets in different postures, calibrate whether their pulling postures are unbalanced, and obtain a training set for training a posture recognition model; The posture recognition model framework is used as a network framework, image sequences of different postures of a person pulling a pet are used as input, and whether the pulling posture is unbalanced is used as a label, so as to train a posture recognition model.

[0011] Optionally, the preset target recognition model is trained by the following steps: Collect multiple images of people towing pets and pedestrians on the road without towing pets as a target recognition training set, and annotate whether the person in each image in the target recognition training set is towing a pet; The convolutional neural network is used as the network framework, the pictures of the target recognition training set are used as input, and whether the pet is being pulled is used as the label, so as to train and obtain the target recognition model.

[0012] Compared with the prior art, the technical solution provided by this application has the following advantages: One of the beneficial effects is that the intelligent watering method provided by the present application uses a camera to obtain an image sequence of pedestrians on a predetermined watering path during the watering process. Then, the image sequence is analyzed using a preset target recognition model to judge whether the pedestrians are pulling a pet. When a pedestrian pulling a pet is detected, the pedestrian pulling the pet is marked as a pulling person. When the watering vehicle approaches the pulling person, the predetermined pre-watering control action is executed in advance, including deceleration, stopping watering or stopping the warning sound, etc., to reduce the interference of the watering operation on the pulled pet, thereby reducing the uncertain behavior of the pet due to stimulation.

[0013] Therefore, the present application reduces the impact of uncertainty caused by pets on the sprinkler operation and pedestrian safety by identifying pedestrians pulling pets and taking sprinkler control actions in advance to reduce the possibility of pet stress. Because the behavior of pets, especially dogs, is uncertain, they are easily stimulated by the external environment, such as the sound of sprinkler vehicles, sprinkler actions or warning signals, which can cause pets to be frightened and pull pedestrians through the traction rope. This may cause pedestrians to accidentally enter the sprinkler area, or lose balance in the process of trying to control the pet, causing non-contact accidents such as falls. The existing technology is only based on the identification of the position of pedestrians, and it is assumed that the behavior of pedestrians is predictable and controllable, and it cannot effectively deal with the behavioral uncertainty caused by pets. The present application fully considers the uncertainty of human behavior caused by the combination of pets and pullers, and by reducing the stimulation of sprinkler operations on pets, it avoids pets from pulling pedestrians into the sprinkler area or causing other safety hazards after being frightened, thereby improving the safety of sprinkler operations.

[0014] Therefore, the road sanitation intelligent sprinkler control method provided in this application can better avoid pedestrians on the road.

[0015] The second beneficial effect is that, since pet dogs have different body sizes, larger dogs have a stronger ability to interfere with their owners, especially when stimulated by external factors, they are likely to pull their owners through the traction rope, causing the owners to lose balance or accidentally enter the watering area. The degree of interference is related to the size of both the pet dog and the owner. When the size of the pet dog approaches or exceeds the size of the owner, the owner's control ability will be further reduced. This application determines the size ratio by considering the size factors of the pet dog and its owner, and then calculates the pre-risk value to obtain the distance at which the watering needs to be controlled in advance, so that the watering vehicle can adjust the control intensity of the watering according to the actual situation. For larger pet dogs, the watering vehicle will start to perform the watering control action in advance at a farther distance, giving the pet a longer time window to adapt, thereby reducing the possibility of the pet producing uncertain behavior and reducing interference with the traction person, thereby preventing possible non-contact accidents. By comprehensively considering the size of the pet and the owner, this application improves the safety and reliability of the watering operation.

[0016] The third beneficial effect is that when calculating the body size ratio, the image sequence of the tractor in the first time window is first obtained. For each image in the first time window, the contours of the tractor and the pet being tracted are extracted respectively to calculate the contour areas of each. Subsequently, the present application uses a weighted average method to comprehensively calculate the body size ratio in each image according to the preset body size ratio calculation function to obtain the final body size ratio, wherein the weight coefficient is determined based on the ratio of the contour area of ​​the tractor pet in each image to the average value of the contour area of ​​the tractor pet in all images in the time window. This calculation method gives higher weights to images with larger contour areas and focuses on the image ratio when the pet presents a larger area, thereby effectively eliminating the area measurement error caused by the tractor and the pet at different angles and postures, and ensuring the stability and accuracy of the body size ratio calculation. In this way, the present application can more accurately evaluate the body size ratio between the tractor and the pet, and then more reasonably determine the pre-risk value, so as to obtain a more reasonable distance to start controlling the watering in advance.

[0017] The fourth beneficial effect is that pets, especially pet dogs, generally have different personalities and training levels. If the range of watering is controlled solely based on the pet's body size, the range of watering stop may be too large, thereby reducing the efficiency of watering operations. To this end, when the watering vehicle approaches the towing person and the pet, the present application further analyzes the visible light image sequence of the towing person through a preset posture recognition model to obtain the posture changes of the towing person, so as to judge the personality, behavior pattern and adaptability of the pet dog, thereby adjusting the subsequent watering stop range.

[0018] When the sprinkler vehicle enters the predetermined pre-sprinkler control action stage, it first enters the interference enhancement stage to enhance the external interference stimulus of the sprinkler, such as increasing the warning volume or frequency. At this time, the posture recognition model is used to monitor the posture changes of the tractor in real time to determine whether there are abnormal conditions such as imbalance caused by the pet's behavior. If it is detected that the tractor's posture is stable and the pet's behavior is stable, the interference enhancement stage will be exited after a predetermined period of time and the interference reduction stage will be entered. According to the duration of the interference enhancement stage, the pre-risk value is adjusted using the preset post-risk value calculation function to obtain the post-risk value. The longer the interference enhancement stage is, the lower the uncertainty of the pet is, the lower the post-risk value is, and the shorter the distance that the sprinkler vehicle needs to continue to control the sprinkler after passing the tractor.

[0019] On the contrary, if the posture of the person pulling the pet is detected to be unstable, indicating that the pet's behavior is unstable, the interference enhancement stage will be immediately exited and the interference reduction stage will be entered. The shorter the duration of the interference enhancement stage, the higher the uncertainty of the pet, and the higher the post-risk value, which will increase the distance of the subsequent control of watering. Through this dynamic adjustment, the sprinkler vehicle can dynamically determine the appropriate stop watering range according to the personality and behavior pattern of the pet dog, thereby maximizing the efficiency of watering. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A flow chart of the road sanitation intelligent sprinkler control method provided in an embodiment of the present application. DETAILED DESCRIPTION

[0021] The technical solution in this application will be described below in conjunction with the accompanying drawings.

[0022] In the following description, many specific details are set forth to facilitate a full understanding of the present application, but the present application can also be implemented in other ways different from those described herein; obviously, the embodiments in the specification are only part of the embodiments of the present application, not all of the embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0023] Reference Figure 1 , the present application provides a road sanitation intelligent sprinkler control method, comprising the following steps: S101: Acquire an image sequence of pedestrians on a predetermined watering path through a camera.

[0024] Specifically, the camera takes photos of the road beside the watering path to be traveled at artificially set intervals, thereby obtaining an image sequence of pedestrians on the road on the predetermined watering path.

[0025] S102: Based on the image sequence of the pedestrians on the road, a preset target recognition model is used to judge whether the pedestrians on the road are pulling the pet.

[0026] Specifically, the preset target recognition model is trained by the following steps: Collect multiple images of people towing pets and pedestrians on the road without towing pets as a target recognition training set, and annotate whether the person in each image in the target recognition training set is towing a pet; The convolutional neural network is used as the network framework, the pictures of the target recognition training set are used as input, and whether the pet is being pulled is used as the label, so as to train and obtain the target recognition model.

[0027] Specifically, the target recognition model is used to perform target recognition on each newly acquired image in the image sequence of pedestrians on the road, so as to distinguish and recognize whether there is a pedestrian towing a pet.

[0028] S103: If there is a pedestrian on the road who is pulling the pet, the pedestrian pulling the pet is marked as the puller, and the distance to the puller to start controlling the watering in advance is calculated.

[0029] Specifically, calculate the distance to the tractor to start watering in advance: Obtaining a body size ratio between the towing person and the towed pet, and calculating a pre-risk value according to the body size ratio by using a preset pre-risk value calculation function, wherein the pre-risk value calculation function is set so that the closer the body size of the towed pet is to the body size of the towing person, the higher the pre-risk value; According to the pre-risk value, the distance to the tractor that needs to be controlled for watering to start in advance is calculated by a preset control watering distance calculation function. The control watering distance calculation function is set so that the larger the value of the pre-risk value, the larger the distance required for controlling watering to start in advance.

[0030] Specifically, in the embodiment of the present application, the pre-risk value calculation function is: in, is the pre-risk value, It is a first empirical parameter preset artificially, and B is the body size ratio of the person pulling and the pet being pulled.

[0031] The control watering distance calculation function is: in, To control the watering distance, is the second empirical parameter preset artificially. is the risk value, which is used to bring in the pre-risk value or post-risk value. Based on the control distance.

[0032] It should be noted that in other embodiments, those skilled in the art should realize that, based on the design concept of the pre-risk value and the control watering distance calculation function, the specific forms of the pre-risk value calculation function and the control watering distance calculation function can have more than one variant, and their variant forms can also achieve the same technical effect. Therefore, the scope of protection requested in this application should not be limited to the specific calculation formula proposed in the embodiments of this application.

[0033] Specifically, the body size ratio of the pulling person and the pulled pet is calculated by the following steps: When the pulling person is detected, an image sequence of the pulling person in a first time window is acquired; For each photo in the first time window, the outlines of the person pulling the pet are extracted.

[0034] After the target recognition model in the present application determines the area where the pulling person and the pulled pet are located, the contours of the pulling person and the pulled pet in the area are extracted using the Canny edge detection algorithm to obtain the contours of the pulling person and the pulled pet.

[0035] In other embodiments, based on further improvement in accuracy, the instance segmentation model Mask R-CNN can be used, but the training of this model requires a large amount of training data. The Mask R-CNN model construction and its training process belong to the prior art, and its training process and model construction are not described in detail here. If this technology is adopted, the target recognition model can be replaced with the Mask R-CNN instance segmentation model, so as to achieve the technical effects of target recognition and contour segmentation at the same time.

[0036] The outline areas of the towing person and the pet being towed in each photo are calculated, and the body size ratio of the towing person and the pet being towed is obtained by a preset body size ratio calculation function.

[0037] There are many existing technologies that can realize the contour area calculation technology. In the embodiment of the present application, a pixel counting method is used to obtain the area of ​​the contour, which will not be described in detail here.

[0038] The preset body shape ratio calculation function is: Where I is the total number of photos in the first time window, i is the i-th photo in the first time window, A(p,i) is the outline area of ​​the pulled pet in the i-th photo, and A(m,i) is the outline area of ​​the pulling person in the i-th photo; in, It is the average of the contour areas of the pulled pet in all photos within the entire first time window.

[0039] S104: Starting to execute a predetermined pre-sprinkling control action in advance when approaching the towing person, so as to reduce the interference of the towed pet on the towing person.

[0040] Specifically, when the sprinkler truck is at a distance from the towing person at which the sprinkler truck needs to start controlling the sprinkler in advance, the sprinkler truck performs a predetermined pre-sprinkling control action.

[0041] The distance between the sprinkler truck and the towing person can be obtained by using a distance sensor on the sprinkler truck or binocular vision to measure the distance, but these are existing technologies in the field and are means that technical personnel in this field can choose according to actual conditions, so they will not be described in detail here.

[0042] The predetermined pre-sprinkler control action includes an interference enhancement stage and an interference reduction stage. When the predetermined pre-sprinkler control action begins to be executed, the interference enhancement stage of a predetermined duration is first entered. In the interference enhancement stage, the interference stimulus to the outside caused by sprinkling is enhanced. After exiting the interference enhancement stage, the interference reduction stage is entered to reduce the interference stimulus to the outside caused by sprinkling until the position of the tractor is reached.

[0043] Specifically, the actions performed in the interference enhancement stage are: increasing the water pressure during watering, increasing the warning volume, and other behaviors to enhance the interference stimulation of watering to the outside.

[0044] Specifically, the actions performed in the interference reduction stage are: reducing the movement speed of the sprinkler truck, stopping watering, turning off the warning sound, and other actions to reduce the interference stimulation of the sprinkler to the outside.

[0045] S105: Calculate the distance that needs to be controlled for watering after the tractor passes by according to the visible light image sequence of the tractor and the pre-risk value when the predetermined pre-watering control action is performed: In the interference enhancement stage, according to the visible light image sequence of the tractor when executing the predetermined pre-sprinkling control action, the posture change of the tractor is obtained through a preset posture recognition model to determine whether to exit the interference enhancement stage at present; According to the length of the interference enhancement phase, the pre-risk value is changed by a preset post-risk value calculation function to obtain a post-risk value, so that the longer the interference enhancement phase is, the lower the post-risk value obtained is; According to the post-risk value, the distance for controlling the watering after the person is pulled is calculated by a preset control watering distance calculation function.

[0046] The post-risk value calculation function is: in, is the post-risk value, For the scheduled duration, The actual duration of the interference enhancement phase, It is an artificially set post-risk value base.

[0047] Based on the design concept of the post-risk value calculation function, the specific form of the post-risk value calculation function can have more than one variant, and its variant form can also achieve the same technical effect. Therefore, the scope of protection requested by this application should not be limited to the specific calculation formula proposed in the embodiment of this application.

[0048] Specifically, the preset posture recognition model is trained by the following steps: Constructing the posture recognition model framework, the posture recognition model framework includes an input layer, a skeleton point feature extraction layer, a feature integration layer, a convolutional neural network layer and an output layer; An input layer, for receiving the visible light image of the traction person after entering the interference enhancement stage; The skeleton point feature extraction layer is a preset skeleton key point extraction algorithm, which is used to receive the data output by the input layer and calibrate the skeleton key points of the tractor in the visible light image of the tractor, and the skeleton key points include shoulders, elbows and legs; Specifically, the skeleton key point extraction algorithm is the OpenPose algorithm or the AlphaPose algorithm. In other embodiments, other posture estimation algorithms can also be used to extract skeleton key points.

[0049] However, it should be noted that even if the embodiment of the present application does not improve the algorithm itself, the embodiment of the present application proposes that technical improvements need to be made to at least the key points of the shoulders, elbows and legs. This is because when a person pulls a pet, the changing angle of the pulling hand and the extent of the body leaning forward and backward are strongly related to whether the human body posture is unbalanced.

[0050] The feature integration layer is used to create a sliding input window of a preset size, which is used to receive the input of the bone point feature extraction layer within a preset size time to form a bone point feature change sequence for external output; This preset size is an artificially set value, usually set to 3-6.

[0051] The convolutional neural network layer is used to receive the skeleton point feature change sequence transmitted by the feature integration layer, perform data processing on the skeleton point feature change sequence and then output it externally; The output layer is used to obtain the output of the convolutional neural network layer, process the output of the convolutional neural network layer and output it externally; Collect image sequences of multiple people pulling pets in different postures, calibrate whether their pulling postures are unbalanced, and obtain a training set for training a posture recognition model; It should be noted that the number of images in the image sequence of different postures here is preferably consistent with the preset size mentioned above.

[0052] The posture recognition model framework is used as a network framework, image sequences of different postures of a person pulling a pet are used as input, and whether the pulling posture is unbalanced is used as a label, so as to train a posture recognition model.

[0053] When the traction posture is determined to be unbalanced, a signal is sent to end the interference enhancement stage, so that the watering behavior immediately enters the interference reduction stage, and then the post-risk value is obtained through the post-risk value calculation function.

[0054] Substitute the obtained post-risk value into the control watering distance calculation function to obtain the distance that needs to be controlled for watering after the person is pulled.

[0055] S106: Based on the obtained distance for controlling watering after the towing person has passed, the watering control action is continued within the distance after the towing person has passed.

[0056] Specifically, the interference reduction stage is maintained until the sprinkler truck moves to a distance beyond which the sprinkler needs to be controlled after passing the towing person, so that the sprinkler truck continues the sprinkler operation.

[0057] Compared with the prior art, the technical solution provided by the embodiments of the present application has the following advantages: One of the beneficial effects is that the intelligent watering method provided by the present application uses a camera to obtain an image sequence of pedestrians on a predetermined watering path during the watering process. Then, the image sequence is analyzed using a preset target recognition model to judge whether the pedestrians are pulling a pet. When a pedestrian pulling a pet is detected, the pedestrian pulling the pet is marked as a pulling person. When the watering vehicle approaches the pulling person, the predetermined pre-watering control action is executed in advance, including deceleration, stopping watering or stopping the warning sound, etc., to reduce the interference of the watering operation on the pulled pet, thereby reducing the uncertain behavior of the pet due to stimulation.

[0058] Therefore, the present application reduces the impact of uncertainty caused by pets on the sprinkler operation and pedestrian safety by identifying pedestrians pulling pets and taking sprinkler control actions in advance to reduce the possibility of pet stress. Because the behavior of pets, especially dogs, is uncertain, they are easily stimulated by the external environment, such as the sound of sprinkler vehicles, sprinkler actions or warning signals, which can cause pets to be frightened and pull pedestrians through the traction rope. This may cause pedestrians to accidentally enter the sprinkler area, or lose balance in the process of trying to control the pet, causing non-contact accidents such as falls. The existing technology is only based on the identification of the position of pedestrians, and it is assumed that the behavior of pedestrians is predictable and controllable, and it cannot effectively deal with the behavioral uncertainty caused by pets. The present application fully considers the uncertainty of human behavior caused by the combination of pets and pullers, and by reducing the stimulation of sprinkler operations on pets, it avoids pets from pulling pedestrians into the sprinkler area or causing other safety hazards after being frightened, thereby improving the safety of sprinkler operations.

[0059] Therefore, the road sanitation intelligent sprinkler control method provided in this application can better avoid pedestrians on the road.

[0060] The second beneficial effect is that, because pet dogs have different body sizes, larger dogs have a stronger ability to interfere with their owners, especially when stimulated by external factors, they are easy to pull their owners through the traction rope, causing the owners to lose balance or accidentally enter the watering area. The degree of interference is related to the size of both the pet dog and the owner. When the size of the pet dog approaches or exceeds the size of the owner, the owner's control ability will be further reduced. This application determines the size ratio by considering the size factors of the pet dog and its owner, and then calculates the pre-risk value to obtain the distance to start controlling the watering in advance, so that the sprinkler vehicle can adjust the range of controlling the watering according to the actual situation. For larger pet dogs, the sprinkler vehicle will start to perform the sprinkler control action in advance at a farther distance, giving the pet a longer time window to adapt, thereby reducing the possibility of the pet producing uncertain behavior and reducing interference with the traction person, thereby preventing possible non-contact accidents. By comprehensively considering the size of the pet and the owner, this application improves the safety and reliability of the sprinkler operation.

[0061] The third beneficial effect is that when calculating the body size ratio, the image sequence of the tractor in the first time window is first obtained. For each image in the first time window, the contours of the tractor and the pet being tracted are extracted respectively to calculate the contour areas of each. Subsequently, the present application uses a weighted average method to comprehensively calculate the body size ratio in each image according to the preset body size ratio calculation function to obtain the final body size ratio, wherein the weight coefficient is determined based on the ratio of the contour area of ​​the tractor pet in each image to the average value of the contour area of ​​the tractor pet in all images in the time window. This calculation method gives higher weights to images with larger contour areas and focuses on the image ratio when the pet presents a larger area, thereby effectively eliminating the area measurement error caused by the tractor and the pet at different angles and postures, and ensuring the stability and accuracy of the body size ratio calculation. In this way, the present application can more accurately evaluate the body size ratio between the tractor and the pet, and then more reasonably determine the pre-risk value, so as to obtain a more reasonable distance to start controlling the watering in advance.

[0062] The fourth beneficial effect is that pets, especially pet dogs, generally have different personalities and training levels. If the distance for controlling the watering is determined based solely on the pet's size, the watering stop range may be too large, thereby reducing the efficiency of the watering operation. To this end, when the watering vehicle approaches the towing person and the pet, the present application further analyzes the towing person's visible light image sequence through a preset posture recognition model to obtain the towing person's posture changes, so as to determine the pet dog's personality, behavior pattern and adaptability, thereby adjusting the subsequent watering stop range.

[0063] When the sprinkler vehicle enters the predetermined pre-sprinkler control action stage, it first enters the interference enhancement stage to enhance the external interference stimulus of the sprinkler, such as increasing the warning volume or frequency. At this time, the posture recognition model is used to monitor the posture changes of the tractor in real time to determine whether there are abnormal conditions such as imbalance caused by the pet's behavior. If it is detected that the tractor's posture is stable and the pet's behavior is stable, the interference enhancement stage will be exited after a predetermined period of time and the interference reduction stage will be entered. According to the duration of the interference enhancement stage, the pre-risk value is adjusted using the preset post-risk value calculation function to obtain the post-risk value. The longer the interference enhancement stage is, the lower the uncertainty of the pet is, the lower the post-risk value is, and the shorter the distance that the sprinkler vehicle needs to stop sprinkling after passing the tractor.

[0064] On the contrary, if the posture of the person pulling the pet is detected to be unstable, indicating that the pet's behavior is unstable, the interference enhancement stage will be immediately exited and the interference reduction stage will be entered. The shorter the duration of the interference enhancement stage, the higher the uncertainty of the pet, and the higher the post-risk value, which will increase the distance of the subsequent control of watering. Through this dynamic adjustment, the sprinkler vehicle can dynamically determine the appropriate stop watering range according to the personality and behavior pattern of the pet dog, thereby maximizing the efficiency of watering.

[0065] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. In addition, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including one..." do not exclude the existence of other identical elements in the process, method, article or device including the elements. Moreover, in the description of the embodiments of the present application, unless otherwise specified, " / " means or, for example, A / B can represent A or B; "and / or" in this article is only a kind of association relationship describing the associated objects, indicating that there can be three relationships, for example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. Furthermore, in the description of the embodiments of the present application, “plurality” refers to two or more than two.

[0066] The above description is only a specific implementation of the present application, so that those skilled in the art can understand or implement the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments described herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A road sanitation intelligent watering control method, characterized in that: The following steps are involved: Acquire an image sequence of pedestrians on a predetermined watering path through a camera; Based on the image sequence of pedestrians on the road, a preset target recognition model is used to judge whether the pedestrians on the road are pulling the pet; If there is a pedestrian on the road pulling his pet, the pedestrian pulling the pet is marked as the puller, and the predetermined pre-sprinkling control action is executed in advance when approaching the puller to reduce the interference of the pulled pet on the puller.

2. The road sanitation intelligent sprinkler control method according to claim 1 is characterized in that: The following steps are also included: Calculate the distance to the tractor to start watering in advance: Obtaining a body size ratio between the towing person and the towed pet, and calculating a pre-risk value according to the body size ratio by using a preset pre-risk value calculation function, wherein the pre-risk value calculation function is set so that the closer the body size of the towed pet is to the body size of the towing person, the higher the pre-risk value; According to the pre-risk value, the distance to the tractor that needs to be controlled for watering in advance is calculated by a preset control watering distance calculation function, and the control watering distance calculation function is set so that the larger the value of the pre-risk value is, the larger the distance that needs to be controlled for watering in advance is; The predetermined pre-watering control action is started at the distance ahead of the towing person at which the watering control needs to be started in advance.

3. The road sanitation intelligent sprinkler control method according to claim 2 is characterized in that: The size ratio of the pulling person and the pulled pet is calculated by the following steps: When the pulling person is detected, an image sequence of the pulling person in a first time window is acquired; For each photo in the first time window, extract the outlines of the person pulling the pet; Calculate the outline area of ​​the towing person and the pet being towed in each photo, and obtain the body size ratio of the towing person and the pet being towed by using a preset body size ratio calculation function; The preset body shape ratio calculation function is: Where I is the total number of photos in the first time window, i is the i-th photo in the first time window, A(p,i) is the outline area of ​​the pulled pet in the i-th photo, and A(m,i) is the outline area of ​​the pulling person in the i-th photo; in, It is the average of the contour areas of the pulled pet in all photos within the entire first time window.

4. The road sanitation intelligent sprinkler control method according to claim 2 is characterized in that: The following steps are also included: The predetermined pre-sprinkler control action includes an interference enhancement phase and an interference reduction phase. When the predetermined pre-sprinkler control action is started, the interference enhancement phase of a predetermined duration is first entered. In the interference enhancement phase, the interference stimulus to the outside during the sprinkling is enhanced. After exiting the interference enhancement phase, the interference reduction phase is entered to reduce the interference stimulus to the outside during the sprinkling until the position of the tractor is reached. According to the visible light image sequence of the tractor when executing the predetermined pre-sprinkling control action and the pre-risk value, the distance to be controlled for sprinkling after the tractor is calculated: In the interference enhancement stage, according to the visible light image sequence of the tractor when executing the predetermined pre-sprinkling control action, the posture change of the tractor is obtained through a preset posture recognition model to determine whether to exit the interference enhancement stage at present; According to the length of the interference enhancement phase, the pre-risk value is changed by a preset post-risk value calculation function to obtain a post-risk value, so that the longer the interference enhancement phase is, the lower the post-risk value obtained is; According to the post-risk value, the distance at which watering needs to be controlled after the person is pulled is calculated by a preset watering control distance calculation function; The interference reduction stage is maintained until the distance after the tractor leaves the tractor and the water sprinkling needs to be controlled, and then the interference reduction stage is exited.

5. The road sanitation intelligent sprinkler control method according to claim 4 is characterized in that: The preset posture recognition model is trained by the following steps: Constructing the posture recognition model framework, the posture recognition model framework includes an input layer, a skeleton point feature extraction layer, a feature integration layer, a convolutional neural network layer and an output layer; An input layer, for receiving the visible light image of the traction person after entering the interference enhancement stage; The skeleton point feature extraction layer is a preset skeleton key point extraction algorithm, which is used to receive the data output by the input layer and calibrate the skeleton key points of the tractor in the visible light image of the tractor, and the skeleton key points include shoulders, elbows and legs; The feature integration layer is used to create a sliding input window of a preset size, which is used to receive the input of the bone point feature extraction layer within a preset size time to form a bone point feature change sequence for external output; The convolutional neural network layer is used to receive the skeleton point feature change sequence transmitted by the feature integration layer, perform data processing on the skeleton point feature change sequence and then output it externally; The output layer is used to obtain the output of the convolutional neural network layer, process the output of the convolutional neural network layer and output it externally; Collect image sequences of multiple people pulling pets in different postures, calibrate whether their pulling postures are unbalanced, and obtain a training set for training a posture recognition model; The posture recognition model framework is used as a network framework, image sequences of different postures of a person pulling a pet are used as input, and whether the pulling posture is unbalanced is used as a label, so as to train a posture recognition model.

6. The road sanitation intelligent sprinkler control method according to claim 1 is characterized in that: The preset target recognition model is trained by the following steps: Collect multiple images of people towing pets and pedestrians on the road without towing pets as a target recognition training set, and annotate whether the person in each image in the target recognition training set is towing a pet; The convolutional neural network is used as the network framework, the pictures of the target recognition training set are used as input, and whether the pet is being pulled is used as the label, so as to train and obtain the target recognition model.

Images