A method for adaptive observation of a patrol robot based on beef cattle abnormal behavior recognition

Abstract

The application discloses a kind of based on beef cattle abnormal behavior identification's inspection robot adaptive observation method, it is related to intelligent breeding, beef cattle abnormal behavior identification and inspection robot control technical field.The method constructs beef cattle abnormal behavior observation knowledge base, the knowledge base stores abnormal behavior type, abnormal behavior identification feature, key observation area, observation strategy parameter, inspection robot constraint parameter and the multistage corresponding relationship between theoretical observation pose.Abnormal behavior type is identified and corresponding key observation area is retrieved in the process of inspection, and theoretical observation pose is generated in combination with the current state of robot, and the final observation pose is generated after being revised by the joint evaluation of image quality, target area integrity and beef cattle stress state, and precise observation is realized through the collaborative control of mobile chassis and pan-tilt, so as to improve the integrity of key area information and observation quality, reduce animal stress reaction, improve autonomous inspection efficiency.

Description

Technical Field

[0001] This invention relates to the fields of intelligent farming, animal behavior recognition, and mobile robot inspection technology, and more specifically to an adaptive observation method for inspection robots based on the recognition of abnormal behavior in beef cattle. Background Technology

[0002] With the development of large-scale beef cattle farming, using inspection robots for cattle health monitoring, abnormal behavior identification, and environmental inspection has become an important technological direction in smart farming. Inspection robots can replace manual labor in continuous patrols of cattle sheds or exercise yards, and acquire cattle behavior data through technologies such as visual perception, environmental perception, and motion control.

[0003] Existing inspection robots typically collect images along fixed routes and using fixed shooting methods, then input the results into an abnormal behavior recognition model. However, the beef cattle farming environment is characterized by its unstructured nature, frequent occlusion, significant changes in lighting, and pronounced autonomous movement and group interaction among cattle. The key observation areas corresponding to different abnormal behaviors are not the same, and a fixed observation pose cannot simultaneously meet the recognition requirements of multiple types of abnormal behaviors.

[0004] For example, drooling behavior mainly relies on fine-grained image information of the oral cavity and mandibular region; lameness requires continuous observation of the movement status of the limbs and joints; abnormal lying behavior requires obtaining the overall posture of the cow; abnormal skin behavior requires stable focusing on the local lesion area; and fighting behavior requires obtaining the multi-target interaction area from a safe distance. If the inspection robot fails to adjust its observation posture based on the abnormal behavior recognition results, problems such as key areas not entering the field of view, partial occlusion, insufficient image clarity, or unreasonable observation distance may easily occur.

[0005] Furthermore, the performance of abnormal behavior recognition models for beef cattle relies on high-quality sample data, while abnormal behavior samples in livestock farming are characterized by low frequency of occurrence, unstable duration, and high difficulty in data collection. If inspection robots can adaptively adjust their observation strategies after identifying suspected abnormal behavior, they can improve the completeness, clarity, and usability of key samples, thereby providing a data foundation for the continuous optimization of abnormal behavior recognition models.

[0006] Meanwhile, when inspection robots approach or follow beef cattle, they may trigger stress responses such as retreating, avoiding, rapid head turns, high-frequency tail swinging, or group agitation. Existing observation methods often focus on image clarity, lacking synergistic constraints on the integrity of key observation areas and the animal's biological stress state, making it difficult to balance observation quality with animal safety.

[0007] Therefore, there is an urgent need to propose an adaptive observation method for beef cattle inspection robots driven by abnormal behavior recognition results, so that the inspection robot can automatically determine key observation areas and generate reasonable observation poses for different abnormal behaviors, and make dynamic corrections in combination with image quality, regional integrity and biological stress state during the observation process. Summary of the Invention

[0008] In view of this, the present invention provides an adaptive observation method for inspection robots based on the recognition of abnormal behavior in beef cattle, which solves the problems of existing inspection robots in the scenario of recognizing abnormal behavior in beef cattle, such as difficulty in obtaining key observation areas for different abnormal behaviors, fixed observation poses, unstable observation quality, and lack of animal stress constraints.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: An adaptive observation method for inspection robots based on the identification of abnormal behavior in beef cattle includes the following steps: A knowledge base for observing abnormal behavior in beef cattle is constructed. This knowledge base stores multi-level correspondences between abnormal behavior identification features, key observation areas, observation strategy parameters, inspection robot parameters, and theoretical observation poses, indexed by abnormal behavior type. The multi-level correspondences include the identification correspondence between abnormal behavior type and abnormal behavior identification features, the regional correspondence between abnormal behavior type and key observation area, the strategy correspondence between key observation area and observation strategy parameters, and the constraint transformation relationship between observation strategy parameters, inspection robot parameters, and theoretical observation poses. During the inspection process, images of the target beef cattle's behavior and spatial location information are acquired. The spatial location information of the target beef cattle includes the coordinates of the target beef cattle in the inspection scene coordinate system, the distance and orientation relative to the inspection robot, and is used to convert key observation areas into executable observation distance, observation direction and gimbal pointing parameters. Based on the target beef cattle behavior images, abnormal behavior of the target beef cattle is identified to obtain the abnormal behavior type and identification confidence level. Then, using the abnormal behavior type and / or abnormal behavior identification features as indexes, the corresponding key observation areas are retrieved in the beef cattle abnormal behavior observation knowledge base. Based on the abnormal behavior type, key observation area, target beef cattle spatial location information, and the current state parameters of the inspection robot, the corresponding observation strategy parameters are called from the beef cattle abnormal behavior observation knowledge base, and a theoretical observation pose is generated under the conditions of satisfying the robot's motion range, gimbal motion range, and safe observation distance constraints. A joint observation and evaluation mechanism is constructed based on the integrity of the target area, image quality, and stress status of beef cattle. The integrity score of the target area, the image quality score, and the stress score of beef cattle are calculated respectively, and the joint observation and evaluation results are output according to the preset weights or thresholds. The theoretical observation pose is adaptively corrected based on the joint observation and evaluation results to generate the final observation pose, and the inspection robot is controlled to collect effective observation data based on the final observation pose.

[0010] Optionally, constructing a knowledge base for observing abnormal behaviors in beef cattle includes: Establish a classification system for abnormal behaviors in beef cattle; Extract visual recognition features corresponding to different abnormal behaviors; Identify the key observation areas corresponding to different abnormal behaviors; Establish a hierarchical index relationship between abnormal behavior types, visual recognition features, key observation areas, and observation strategy parameters; Based on the key observation area, the observation angle, observation distance, observation height, observation duration, gimbal attitude parameters, and motion following mode are generated. The theoretical observation pose is generated by combining the size of the inspection robot, the installation height of the camera, the range of motion of the gimbal, the movement speed constraints, and the safe observation distance. Establish a mapping relationship between abnormal behavior types, key observation areas, spatial location information of target beef cattle, current state parameters of inspection robots and theoretical observation poses, and form a knowledge base for observing abnormal behavior of beef cattle.

[0011] Optionally, the abnormal behavior includes one or more of the following: drooling, lameness, abnormal lying down, abnormal skin behavior, and fighting behavior; Abnormal behavior identification features include: The characteristics of drooling behavior, including the characteristics of liquid strands hanging from the corners of the mouth, the characteristics of liquid droplets on the mandible, and the characteristics of continuous drooling; The characteristics of gait asymmetry, abnormal limb weight-bearing, and abnormal joint movement corresponding to lameness; The characteristics of persistent lying down, difficulty getting up and lying down, and abnormal posture corresponding to abnormal lying behavior; Characteristics of abnormal skin behavior, including hair loss, redness and swelling, scabbing, and trauma. The head collision characteristics, chasing characteristics, and group aggregation characteristics corresponding to fighting behavior.

[0012] Optionally, the current status parameters of the inspection robot include one or more of the following: robot position, robot posture, moving speed, camera position, gimbal pitch angle, gimbal yaw angle, and the relative distance between the robot and the target cattle.

[0013] Optionally, the theoretical observation pose includes one or more of the following: observation distance, observation direction, observation height, observation duration, target observation area, gimbal pitch angle, gimbal yaw angle, and motion following mode, and the theoretical observation pose satisfies the constraints of the inspection robot's motion range, gimbal motion range, and safe observation distance.

[0014] Optionally, different anomalous behaviors correspond to different theoretical observation poses, including: When drooling behavior is detected, a forward observation pose facing the oral cavity and mandibular region of the target beef cattle is generated; When lameness is detected, a lateral following observation pose is generated facing the limbs and joint areas of the target beef cattle; When abnormal lying behavior is detected, a side-above observation pose is generated facing the overall posture area of ​​the target beef cattle; When abnormal skin behavior is detected, a localized focused observation pose is generated for the target lesion area; When fighting behavior is detected, a long-range wide-angle observation pose is generated for the multi-target interaction area.

[0015] Optionally, the joint observation and evaluation results include image quality evaluation results, target area integrity evaluation results, and beef cattle biological stress state evaluation results; wherein, the image quality evaluation results are obtained by weighting image sharpness, brightness, and occlusion degree; the target area integrity evaluation results are obtained by the coverage of key observation areas in the image and the offset of the field of view center; the beef cattle biological stress state evaluation results are obtained by weighting one or more of the following: target beef cattle retreating behavior, continuous avoidance behavior, rapid head turning behavior, high-frequency tail swinging behavior, sudden change in gait frequency behavior, and abnormal fluctuations in herd movement; the joint observation and evaluation results are obtained by weighted fusion or threshold determination of the image quality evaluation results, target area integrity evaluation results, and beef cattle biological stress state evaluation results.

[0016] Optionally, adaptively correcting the theoretical observation pose based on the joint observation evaluation results includes: When the image quality assessment result is lower than the first preset threshold, adjust the observation distance, gimbal pitch angle, or gimbal yaw angle. When the target area integrity assessment result is lower than the second preset threshold, adjust the observation direction, observation angle, or motion following mode. When the assessment result of the biological stress state of beef cattle is higher than the third preset threshold, increase the observation distance, reduce the movement speed, change the observation direction, or perform avoidance actions. The final observation pose is generated based on the above adjustments.

[0017] Optionally, controlling the mobile chassis and gimbal camera to perform the observation task includes: planning the movement path of the mobile chassis according to the final observation pose, controlling the mobile chassis to move to a position that meets the safe observation distance, and controlling the gimbal camera to aim at the key observation area for continuous or intermittent observation.

[0018] Optionally, after collecting valid observation data, the final observation pose, joint observation evaluation results, and abnormal behavior identification results are fed back to the beef cattle abnormal behavior observation knowledge base to update the observation strategy parameters and improve the observation adaptability in subsequent abnormal behavior identification scenarios.

[0019] As can be seen from the above technical solution, compared with the prior art, the present invention provides an adaptive observation method for inspection robots based on the identification of abnormal behavior in beef cattle, which has the following beneficial effects: First, the present invention uses the results of abnormal behavior identification as the direct basis for observation decisions, enabling the inspection robot to automatically select key observation areas and observation methods for different abnormal behaviors. Second, this invention enables the rapid generation of theoretical observation poses through a knowledge base for observing abnormal behavior in beef cattle, thereby improving the targeting of inspection and observation. Third, this invention achieves dynamic correction of observation pose through joint evaluation of image quality, target area integrity, and biological stress state, thereby reducing interference to beef cattle while ensuring complete acquisition of key areas. Fourth, this invention can generate effective observation data feedback, improve the efficiency of abnormal behavior sample collection and the subsequent optimization capability of the identification model. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0021] Figure 1 The figure shown is a schematic diagram of the overall process of an adaptive observation method for inspection robots based on a knowledge base for observing abnormal behavior of beef cattle, according to an embodiment of the present invention. Figure 2 The diagram shown is a schematic diagram of an inspection robot structure according to an embodiment of the present invention; Figure 3 The diagram shown is a schematic diagram of the construction process of a knowledge base for observing abnormal behavior in beef cattle according to an embodiment of the present invention; Figure 4 The diagram shown is a schematic representation of the structural relationship of a knowledge base for observing abnormal behavior in beef cattle according to an embodiment of the present invention. Figure 5 The diagram shown is a schematic diagram of a theoretical observation pose generation process provided by an embodiment of the present invention; Figure 6 The diagram shown is a schematic representation of a joint observation and evaluation process provided in an embodiment of the present invention. Figure 7 The diagram shown is a schematic diagram of an adaptive correction process for observation pose provided by an embodiment of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Example 1

[0024] This embodiment discloses an adaptive observation method for a cattle inspection robot based on the identification of abnormal behavior in beef cattle. The method is applied to a beef cattle inspection robot including a mobile chassis, a gimbal camera device, a visual perception device, and an environmental perception module. The method includes: A knowledge base for observing abnormal behavior in beef cattle is constructed. This knowledge base stores multi-level correspondences between abnormal behavior identification features, key observation areas, observation strategy parameters, inspection robot parameters, and theoretical observation poses, indexed by abnormal behavior type. The multi-level correspondences include the identification correspondence between abnormal behavior type and abnormal behavior identification features, the regional correspondence between abnormal behavior type and key observation area, the strategy correspondence between key observation area and observation strategy parameters, and the constraint transformation relationship between observation strategy parameters, inspection robot parameters, and theoretical observation poses. During the inspection, the visual perception device acquires images of the target beef cattle's behavior, and the environmental perception module acquires the target beef cattle's spatial position information. The target beef cattle's spatial position information is used to determine the target beef cattle's distance, orientation, and inspection scene coordinates relative to the inspection robot, and serves as a spatial constraint for generating theoretical observation pose and correcting final observation pose. Based on the target beef cattle behavior images, abnormal behavior of the target beef cattle is identified to obtain the abnormal behavior type and identification confidence level. Then, using the abnormal behavior type and / or abnormal behavior identification features as indexes, the corresponding key observation areas are retrieved in the beef cattle abnormal behavior observation knowledge base. Based on the abnormal behavior type, key observation area, target beef cattle spatial location information, and the current state parameters of the inspection robot, the corresponding observation strategy parameters are called from the beef cattle abnormal behavior observation knowledge base, and a theoretical observation pose is generated under the conditions of satisfying the robot's motion range, gimbal motion range, and safe observation distance constraints. A joint observation and evaluation mechanism is constructed based on the integrity of the target area, image quality, and stress status of beef cattle. The integrity score of the target area, the image quality score, and the stress score of beef cattle are calculated respectively, and the joint observation and evaluation results are output according to the preset weights or thresholds. The theoretical observation pose is adaptively corrected based on the joint observation and evaluation results to generate the final observation pose. Based on the final observation pose, control the mobile chassis and gimbal camera device to perform the observation task and collect valid observation data.

[0025] Furthermore, the construction of the knowledge base for observing abnormal behavior in beef cattle includes: Establish a classification system for abnormal behaviors in beef cattle; Extract visual recognition features corresponding to different abnormal behaviors; Identify the key observation areas corresponding to different abnormal behaviors; Establish a hierarchical index relationship between abnormal behavior types, visual recognition features, key observation areas, and observation strategy parameters; Based on the key observation area, the observation angle, observation distance, observation height, observation duration, gimbal attitude parameters, and motion following mode are generated. The theoretical observation pose is generated by combining the size of the inspection robot, the installation height of the camera, the range of motion of the gimbal, the movement speed constraints, and the safe observation distance. Establish a mapping relationship between abnormal behavior types, key observation areas, spatial location information of target beef cattle, current state parameters of inspection robots and theoretical observation poses, and form a knowledge base for observing abnormal behavior of beef cattle.

[0026] Furthermore, the abnormal behavior includes one or more of the following: drooling, lameness, abnormal lying down, abnormal skin behavior, and fighting behavior.

[0027] Furthermore, the abnormal behavior identification features include: The characteristics of drooling behavior, including the characteristics of liquid strands hanging from the corners of the mouth, the characteristics of liquid droplets on the mandible, and the characteristics of continuous drooling; The characteristics of gait asymmetry, abnormal limb weight-bearing, and abnormal joint movement corresponding to lameness; The characteristics of persistent lying down, difficulty getting up and lying down, and abnormal posture corresponding to abnormal lying behavior; Characteristics of abnormal skin behavior, including hair loss, redness and swelling, scabbing, and trauma. The head collision characteristics, chasing characteristics, and group aggregation characteristics corresponding to fighting behavior.

[0028] Furthermore, the current status parameters of the inspection robot include one or more of the following: robot position, robot posture, moving speed, camera position, gimbal pitch angle, gimbal yaw angle, and the relative distance between the robot and the target cattle.

[0029] Furthermore, the theoretical observation pose includes one or more of the following: observation distance, observation direction, observation height, observation duration, target observation area, gimbal pitch angle, gimbal yaw angle, and motion following mode. The theoretical observation pose also satisfies the constraints of the inspection robot's motion range, gimbal motion range, and safe observation distance.

[0030] Furthermore, different anomalous behaviors correspond to different theoretical observation poses, including: When drooling behavior is detected, a forward observation pose facing the oral cavity and mandibular region of the target beef cattle is generated; When lameness is detected, a lateral following observation pose is generated facing the limbs and joint areas of the target beef cattle; When abnormal lying behavior is detected, a side-above observation pose is generated facing the overall posture area of ​​the target beef cattle; When abnormal skin behavior is detected, a localized focused observation pose is generated for the target lesion area; When fighting behavior is detected, a long-range wide-angle observation pose is generated for the multi-target interaction area.

[0031] Furthermore, the joint observation and evaluation results include image quality evaluation results, target area integrity evaluation results, and beef cattle biological stress status evaluation results; wherein, the image quality evaluation results are obtained by weighting image sharpness, brightness, and occlusion degree, and the target area integrity evaluation results are obtained by the coverage of key observation areas in the image and the offset of the field of view center.

[0032] Furthermore, the assessment results of the biological stress state of the beef cattle are obtained based on one or more of the following: backward behavior, continuous avoidance behavior, rapid head turning behavior, high-frequency tail swinging behavior, sudden change in gait frequency, and abnormal fluctuations in group movement.

[0033] Furthermore, adaptive correction of the theoretical observation pose based on the joint observation evaluation results includes: When the image quality assessment result is lower than the first preset threshold, adjust the observation distance, gimbal pitch angle, or gimbal yaw angle. When the target area integrity assessment result is lower than the second preset threshold, adjust the observation direction, observation angle, or motion following mode. When the assessment result of the biological stress state of beef cattle is higher than the third preset threshold, increase the observation distance, reduce the movement speed, change the observation direction, or perform avoidance actions. The final observation pose is generated based on the above adjustments.

[0034] Furthermore, controlling the mobile chassis and gimbal camera to perform observation tasks includes: planning the movement path of the mobile chassis according to the final observation pose, controlling the mobile chassis to move to a position that meets the safe observation distance, and controlling the gimbal camera to aim at the key observation area for continuous or intermittent observation.

[0035] Furthermore, after collecting valid observation data, the final observation pose, joint observation evaluation results, and abnormal behavior identification results are fed back to the beef cattle abnormal behavior observation knowledge base to update the observation strategy parameters and improve the observation adaptability in subsequent abnormal behavior identification scenarios.

[0036] Example 2

[0037] This embodiment, in conjunction with the accompanying drawings, further illustrates the present invention: like Figure 1 As shown, this invention provides an adaptive observation method for inspection robots based on the identification of abnormal behavior in beef cattle, comprising the following steps: S100: Construct a knowledge base for observing abnormal behaviors in beef cattle; S200: Identify abnormal behavior in target beef cattle; S300: Generates theoretical observation poses based on anomalous behavior types; S400: Perform joint observation and assessment; S500: Adaptive correction of the observed pose based on the evaluation results; S600: Controls the inspection robot to perform observation tasks. If the preset conditions are not met, it returns to S400 and re-executes the joint observation and evaluation.

[0038] The inspection robot includes a mobile chassis, a gimbal camera device, a binocular camera, a lidar, and a controller, such as... Figure 2 As shown.

[0039] The abnormal behaviors of beef cattle include one or more of the following: drooling, lameness, abnormal lying down, abnormal skin behavior, and fighting behavior.

[0040] like Figure 3 As shown, step S100 further includes: S110: Establish a classification system for abnormal behavior; S120: Extract visual features of abnormal behavior; S130: Identify key observation areas; S140: Generate observation strategy; S150: Integrated inspection robot parameters; S160: Construct an abnormal behavior observation knowledge base M300.

[0041] Specifically: For drooling behavior, the key observation areas are the oral cavity and mandibular region; For lameness, the key observation area is determined to be the limbs and joints. For abnormal lying-down behavior, the key observation area was determined to be the overall posture area of ​​the cattle's body. For abnormal skin behavior, the key observation area is identified as the lesion area; For fighting behavior, the key observation area is determined to be the multi-target interaction area.

[0042] The correspondence of key observation areas is determined based on the visual representation of abnormal behavior: abnormal representation of drooling behavior is concentrated in the oral cavity and mandible; abnormal representation of lameness behavior is concentrated in the weight-bearing of the limbs and joint movement; abnormal representation of lying behavior is concentrated in the bovine posture; abnormal representation of skin behavior is concentrated in lesions on the body surface; and abnormal representation of fighting behavior is concentrated in the areas of head collisions, chasing, and spatial interaction among multiple individuals.

[0043] The observation strategy includes observation angle, observation distance, observation height, observation duration, gimbal attitude parameters, and motion following mode.

[0044] The parameters of the inspection robot include robot size parameters, camera installation height parameters, gimbal movement range parameters, movement speed parameters, and safe observation distance parameters.

[0045] By establishing a chain mapping relationship of "abnormal behavior type - visual recognition feature - key observation area - observation strategy parameter - robot parameter constraint - theoretical observation pose", an abnormal behavior observation knowledge base M300 is formed. Each abnormal behavior type corresponds to at least one visual recognition feature and at least one key observation area. Each key observation area corresponds to at least one set of observation strategy parameters. The observation strategy parameters are converted into theoretical observation pose under constraints of robot size, camera installation height, gimbal motion range, movement speed, and safe observation distance.

[0046] For example, drooling corresponds to the characteristics of liquid hanging from the corners of the mouth, the characteristics of liquid droplets on the mandible, and the oral cavity and mandibular region, and corresponds to the anterior close-range fixed-point observation strategy; limping corresponds to the characteristics of gait asymmetry, abnormal limb weight-bearing, and the limbs and joint regions, and corresponds to the lateral continuous following observation strategy; abnormal lying behavior corresponds to the characteristics of continuous lying down, difficulty in getting up and down, and the overall posture region, and corresponds to the lateral upper overall observation strategy; abnormal skin behavior corresponds to the characteristics of hair loss, redness, swelling, scabbing or trauma, and the lesion region, and corresponds to the local focused observation strategy; fighting behavior corresponds to the characteristics of head collision, chasing and group gathering, and the multi-target interaction region, and corresponds to the long-distance wide-angle observation strategy.

[0047] like Figure 4 and Figure 5 As shown, step S300 includes: S310: Obtain the type of abnormal behavior; S320: Obtain the current status parameters of the inspection robot; S330: Invoke the abnormal behavior observation knowledge base M300; S340: Generates the theoretical observation pose D340.

[0048] When generating the theoretical observation pose D340, the relative distance and azimuth of the target beef cattle relative to the inspection robot are determined by the spatial position information of the target beef cattle, the target pointing of the camera device is determined by the key observation area, and the reachable motion range is determined by the current state parameters of the inspection robot; thereby, the observation strategy parameters in the knowledge base are converted into the target position of the mobile chassis and the target posture of the gimbal.

[0049] in: When drooling is detected, an anterior observation pose facing the oral cavity and mandibular region is generated; When limping behavior is detected, a lateral following observation pose facing the limbs and joint areas is generated; When reclining abnormal behavior is detected, a side-above observation pose facing the overall posture area is generated; When abnormal skin behavior is detected, a close-focused observation pose facing the lesion area is generated; When fighting behavior is detected, a long-distance wide-angle observation pose is generated for the interactive area.

[0050] The theoretical observation pose D340 includes: Observation distance; Observation direction; Observational altitude; Gimbal tilt angle; Gimbal yaw angle; Movement following method.

[0051] like Figure 6As shown, step S400 includes: S410: Perform image quality assessment; S420: Perform a target area integrity assessment; S430: Perform a biological stress assessment; S440: Perform comprehensive observation and assessment.

[0052] in: Step S410 outputs image quality assessment result D411; Step S420 outputs the target area integrity assessment result D421; Step S430 outputs the biological stress assessment result D431.

[0053] Image quality assessment uses the Laplacian operator to calculate the image sharpness index, and combines brightness uniformity and occlusion ratio to obtain the image quality assessment result D411; The target region integrity assessment is calculated by the coverage of the key observation region in the image and the offset between the center of the key observation region and the center of the image, and the target region integrity assessment result D421 is obtained. Biological stress status assessment is determined by the target beef cattle's backward movement, continuous avoidance behavior, rapid head turning behavior, high-frequency tail swinging behavior, sudden changes in gait frequency, and abnormal fluctuations in group movement, and the biological stress assessment result D431 is obtained.

[0054] Step S440 performs normalization and fusion analysis on D411, D421, and D431 to generate a comprehensive evaluation result; specifically, after normalizing D411, D421, and D431 to the same dimension, the comprehensive score is calculated according to E=w1D411+w2D421-w3D431, or compared with preset thresholds to generate a joint observation evaluation result, where w1, w2, and w3 are preset weights.

[0055] like Figure 7 As shown, step S500 includes: Input the theoretical observed pose D340 and the evaluation results D411, D421 and D431 into the pose correction module M500; S510: Perform sharpness correction; S520: Perform integrity correction; S530: Implement stress correction; Output the final observed pose D500.

[0056] in: When the image quality does not meet the threshold requirements, the image clarity is improved by adjusting the observation distance and the gimbal attitude. When the integrity of the target area does not meet the threshold requirements, the visibility of the target area is improved by adjusting the observation azimuth and observation angle. When biological stress is detected, interference with the target cattle can be reduced by increasing the observation distance, decreasing the movement speed, or adjusting the observation direction.

[0057] The corrected effective observation parameters can be fed back to the anomalous behavior observation knowledge base M300 to update the observation strategies in the knowledge base.

[0058] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0059] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An adaptive observation method for inspection robots based on the recognition of abnormal behavior in beef cattle, characterized in that, Includes the following steps: A knowledge base for observing abnormal behavior in beef cattle is constructed. This knowledge base stores multi-level correspondences between abnormal behavior identification features, key observation areas, observation strategy parameters, inspection robot parameters, and theoretical observation poses, indexed by abnormal behavior type. The multi-level correspondences include the identification correspondence between abnormal behavior type and abnormal behavior identification features, the regional correspondence between abnormal behavior type and key observation area, the strategy correspondence between key observation area and observation strategy parameters, and the constraint transformation relationship between observation strategy parameters, inspection robot parameters, and theoretical observation poses. During the inspection process, images of the target beef cattle's behavior and spatial location information are acquired. The spatial location information of the target beef cattle includes the coordinates of the target beef cattle in the inspection scene coordinate system, the distance and orientation relative to the inspection robot, and is used to convert key observation areas into executable observation distance, observation direction and gimbal pointing parameters. Based on the target beef cattle behavior images, abnormal behavior of the target beef cattle is identified to obtain the abnormal behavior type and identification confidence level. Then, using the abnormal behavior type and / or abnormal behavior identification features as indexes, the corresponding key observation areas are retrieved in the beef cattle abnormal behavior observation knowledge base. Based on the abnormal behavior type, key observation area, target beef cattle spatial location information, and the current state parameters of the inspection robot, the corresponding observation strategy parameters are called from the beef cattle abnormal behavior observation knowledge base, and a theoretical observation pose is generated under the conditions of satisfying the robot's motion range, gimbal motion range, and safe observation distance constraints. A joint observation and evaluation mechanism is constructed based on the integrity of the target area, image quality, and stress status of beef cattle. The integrity score of the target area, the image quality score, and the stress score of beef cattle are calculated respectively, and the joint observation and evaluation results are output according to the preset weights or thresholds. The theoretical observation pose is adaptively corrected based on the joint observation and evaluation results to generate the final observation pose, and the inspection robot is controlled to collect effective observation data based on the final observation pose.

2. The adaptive observation method for inspection robots based on abnormal behavior recognition of beef cattle according to claim 1, characterized in that, The construction of a knowledge base for observing abnormal behaviors in beef cattle includes: Establish a classification system for abnormal behaviors in beef cattle; Extract visual recognition features corresponding to different abnormal behaviors; Identify the key observation areas corresponding to different abnormal behaviors; Establish a hierarchical index relationship between abnormal behavior types, visual recognition features, key observation areas, and observation strategy parameters; Based on the key observation area, the observation angle, observation distance, observation height, observation duration, gimbal attitude parameters, and motion following mode are generated; The theoretical observation pose is generated by combining the size of the inspection robot, the installation height of the camera, the range of motion of the gimbal, the speed constraints, and the safe observation distance. Establish a mapping relationship between abnormal behavior types, key observation areas, spatial location information of target beef cattle, current state parameters of inspection robots, and theoretical observation poses to form a knowledge base for observing abnormal behavior of beef cattle.

3. The adaptive observation method for inspection robots based on abnormal behavior recognition of beef cattle according to claim 1, characterized in that, The abnormal behaviors include one or more of the following: drooling, lameness, abnormal lying down, abnormal skin behaviors, and fighting behaviors. Abnormal behavior identification features include: The characteristics of drooling behavior, including the characteristics of liquid hanging from the corners of the mouth, the characteristics of liquid droplets on the mandible, and the characteristics of continuous drooling; The characteristics of gait asymmetry, abnormal limb weight-bearing, and abnormal joint movement corresponding to lameness; The characteristics of persistent lying down, difficulty getting up and lying down, and abnormal posture corresponding to abnormal lying behavior; Characteristics of abnormal skin behavior, including hair loss, redness and swelling, scabbing, and trauma. The head collision characteristics, chasing characteristics, and group aggregation characteristics corresponding to fighting behavior.

4. The adaptive observation method for inspection robots based on abnormal behavior recognition of beef cattle according to claim 1, characterized in that, The current status parameters of the inspection robot include one or more of the following: robot position, robot posture, moving speed, camera position, gimbal pitch angle, gimbal yaw angle, and the relative distance between the robot and the target cattle.

5. The adaptive observation method for inspection robots based on abnormal behavior recognition of beef cattle according to claim 1, characterized in that, The theoretical observation pose includes one or more of the following: observation distance, observation direction, observation height, observation duration, target observation area, gimbal pitch angle, gimbal yaw angle, and motion following mode. The theoretical observation pose also satisfies the constraints of the inspection robot's motion range, gimbal motion range, and safe observation distance.

6. The adaptive observation method for inspection robots based on abnormal behavior recognition of beef cattle according to claim 1, characterized in that, Different anomalous behaviors correspond to different theoretical observation poses, including: When drooling behavior is detected, a forward observation pose facing the oral cavity and mandibular region of the target beef cattle is generated; When lameness is detected, a lateral following observation pose is generated facing the limbs and joint areas of the target beef cattle; When abnormal lying behavior is detected, a side-above observation pose is generated facing the overall posture area of ​​the target beef cattle; When abnormal skin behavior is detected, a localized focused observation pose is generated for the target lesion area; When fighting behavior is detected, a long-range wide-angle observation pose is generated for the multi-target interaction area.

7. The adaptive observation method for inspection robots based on abnormal behavior recognition of beef cattle according to claim 1, characterized in that, The joint observation and evaluation results include image quality evaluation results, target area integrity evaluation results, and beef cattle biological stress state evaluation results. The image quality evaluation results are obtained by weighting image sharpness, brightness, and occlusion degree; the target area integrity evaluation results are obtained by the coverage of key observation areas in the image and the offset of the field of view center; and the beef cattle biological stress state evaluation results are obtained by weighting one or more of the following: target beef cattle retreating behavior, continuous avoidance behavior, rapid head turning behavior, high-frequency tail swaying behavior, sudden changes in gait frequency, and abnormal fluctuations in herd movement. The joint observation and evaluation results are obtained by weighted fusion or threshold determination of the image quality evaluation results, target area integrity evaluation results, and beef cattle biological stress state evaluation results.

8. The adaptive observation method for inspection robots based on abnormal behavior recognition of beef cattle according to claim 1, characterized in that, Adaptive correction of the theoretical observation pose based on the joint observation evaluation results includes: When the image quality assessment result is lower than the first preset threshold, adjust the observation distance, gimbal pitch angle, or gimbal yaw angle. When the target area integrity assessment result is lower than the second preset threshold, adjust the observation direction, observation angle, or motion following mode. When the assessment result of the biological stress state of beef cattle is higher than the third preset threshold, increase the observation distance, reduce the movement speed, change the observation direction, or perform avoidance actions. The final observation pose is generated based on the above adjustments.

9. The adaptive observation method for inspection robots based on abnormal behavior recognition of beef cattle according to claim 1, characterized in that, Controlling the mobile chassis and gimbal camera to perform observation tasks includes: planning the movement path of the mobile chassis according to the final observation pose, controlling the mobile chassis to move to a position that meets the safe observation distance, and controlling the gimbal camera to aim at the key observation area for continuous or intermittent observation.

10. The adaptive observation method for inspection robots based on abnormal behavior recognition of beef cattle according to claim 1, characterized in that, After collecting valid observation data, the final observation pose, joint observation evaluation results, and abnormal behavior identification results are fed back to the beef cattle abnormal behavior observation knowledge base to update the observation strategy parameters and improve the observation adaptability in subsequent abnormal behavior identification scenarios.

Images