CN-121994229-A - Method for predicting obstacle avoidance and optimization of trend path of inspection robot dynamic target

CN121994229ACN 121994229 ACN121994229 ACN 121994229ACN-121994229-A

Abstract

The invention discloses a method for predicting obstacle avoidance and trend path optimization of a dynamic target of a patrol robot, which comprises the steps of loading a built static environment grid map by the patrol robot, matching laser radar scanning data with the static environment grid map, acquiring a thermal image and a depth image of a field environment by the patrol robot, registering the depth image and the thermal image based on a mutual information algorithm, carrying out global path planning on the patrol robot, identifying potential collision points based on a track interaction method based on a prediction perception global path re-planning strategy, triggering global path dynamic re-planning, guiding the robot to avoid a dynamic target in advance, integrating real-time temperature perception information when the global path is safely executed, carrying out quantitative perception on the high-temperature target, establishing a multi-layer temperature intensity radiation model based on a space gradient field, simulating a thermal diffusion effect dynamic optimization local track of the high-temperature target in a real scene, and guiding the patrol robot to autonomously face a high-temperature region.

Inventors

LIU YI
ZHAO ZIYU
SANG HONG
DONG HANG

Assignees

大连海事大学

Dates

Publication Date: 20260508
Application Date: 20251230

Claims (4)

1. A method for predicting obstacle avoidance and trend path optimization of a patrol robot dynamic target is characterized by comprising the following steps: The inspection robot loads the constructed static environment grid map, and adopts laser radar scanning data to match with the static environment grid map so as to perform global map positioning on the inspection robot and determine initial pose information, wherein the static environment grid map comprises fixed layout information of an inspection area; The inspection robot acquires a thermal image and a depth image of the field environment, the thermal image is segmented and binarized by adopting an entropy threshold method, and the maximum communication area is used as a high-temperature dangerous area; registering the depth image and the thermal image based on a mutual information algorithm, mapping pixel coordinates of a high-temperature target in the thermal image into the depth image through a registration relation, and obtaining three-dimensional coordinates of the high-temperature target in the depth image; Carrying out global path planning on the inspection robot by adopting a global path re-planning strategy based on predictive perception, identifying potential collision points based on a track interaction method, triggering global path dynamic re-planning, and guiding the robot to avoid dynamic targets in advance; When the global path is safely executed, the inspection robot integrates real-time temperature sensing information, quantitatively senses a high-temperature target, establishes a multi-layer temperature intensity radiation model based on a spatial gradient field, simulates a thermal diffusion effect of the high-temperature target in a real scene to dynamically optimize a local track, and guides the inspection robot to automatically face a high-temperature area.
2. The method for predicting obstacle avoidance and trend path optimization of the inspection robot dynamic target according to claim 1, wherein in global path planning of the inspection robot, YOLOv is adopted to detect pedestrian targets, deepSORT is utilized to continuously track and smooth the target frame by frame, macroscopic motion displacement vectors of the inspection robot are calculated based on historical track data in a section of observation window, meanwhile, an adaptive correction function aiming at vertical displacement is introduced, and short-term positions of the dynamic targets are predicted based on the correction displacement vectors, specifically, the method comprises the following steps: Let i be the person whose trajectory pi is a list of a series of position coordinates: Wherein, the Representing the coordinates of the pedestrian i at time t, continuously updating the position data of the pedestrian to form a series of continuous track points To capture pedestrian motion trends, define the total displacement vector within the observation window [0, t ]: The displacement vector characterizes the overall motion trend of the pedestrian i in a T frame, and in the track prediction stage, a moving path diagram of the pedestrian i is constructed by analyzing historical moving data delta Pi of the pedestrian i from the first occurrence to the current position, and an adaptive correction function is introduced in consideration of non-uniformity of the motion of the pedestrian: In a navigation scene of the inspection robot, an observation object is mainly an oncoming pedestrian moving along the y-axis direction of an image, and an adaptive correction function and a correction coefficient function are introduced to the vertical direction ( ) The definition is as follows: When (when) When the pedestrian is indicated to be approaching quickly, the pedestrian is based on the corrected displacement vector Predicting a future movement track of the pedestrian: Wherein, the , And (3) representing the current time position coordinate, wherein k represents the prediction step number, k=1 represents the next time, T represents the total frame number of the observation window, and the prediction range satisfies k less than or equal to T, so that the short-term prediction reliability is ensured.
3. The method for optimizing obstacle avoidance and surveillance path of a patrol robot according to claim 1, wherein when a potential collision point is identified based on a track interaction method, the approach degree between a robot moving path and a pedestrian predicted track is determined according to a set distance threshold delta Wherein the positions of the robot and the pedestrian are respectively represented by vectors And In two-dimensional space, when the Euclidean distance d between the two is smaller than the threshold value, the time when the two reach the intersection point i is estimated and is recorded as tr and tp Vr and vp are speeds of the robot and the pedestrian respectively, speeds and current positions of the pedestrian and the robot are analyzed, arrival time is compared with a set time window tau, and by comparing time difference of arrival intersection points of the pedestrian and the robot, sufficient reaction time of the robot is ensured to carry out obstacle avoidance strategy and path adjustment 。
4. The method for molding a semi-closed cavity of a medium-and large-sized thick-wall steel casting according to claim 3, wherein the multi-layer temperature intensity radiation model adopts an exponentially increasing concentric ring structure, and the radius ri of the i-th layer is as follows: Where r 0 is the reference radius, λ=1.1 is the growth coefficient, defining the temperature intensity gradient field: ti is the temperature intensity value of the ith layer, tc is the core temperature intensity, namely the temperature intensity of a high-temperature target, deltaT is the intensity attenuation span, and n is the total layer number of the model; Within the annular region Ω covered by the model, dense discrete point sampling is performed: for the sampling point set P, a temperature intensity visualization map is established, and each sampling point pi is associated with a temperature intensity value Ti to form a set of point sets with temperature intensity labels, which cover the real high-temperature heat source peripheral space For candidate trajectories The dangerous situation environment perception evaluation sub-function is defined as: Wherein, the The method is characterized in that the method is a track end point pose, wherein the evaluation function only uses the end point pose of the candidate track to calculate the distance to the sampling point, and the speed v and the angular speed omega are not directly introduced.

Description

Method for predicting obstacle avoidance and optimization of trend path of inspection robot dynamic target Technical Field The invention relates to the field of intelligent control and navigation of robots, in particular to a dynamic obstacle avoidance and path planning method applied to a patrol robot in a complex industrial environment. Background In industrial scenes such as warehouse, transformer substation, etc., fire hazards pose a serious threat to safety and property. The traditional manual inspection method has the problems of high labor cost, easy error, difficult operation in dangerous environments and the like. With advances in industrial automation and robotics, inspection robots equipped with vision and thermal imaging sensors have become reliable solutions to improve operational safety and efficiency. However, in unstructured dynamic environments, the motion of dynamic objects (e.g., pedestrians) makes the inspection robot appear to be unreliable in obstacle avoidance and real-time path planning delays in the dynamic environment. In addition, in a high-temperature dynamic scene, the inspection robot faces unique challenges of not only effectively avoiding dynamic targets, but also coping with potential fire risks in a high-temperature area. Existing patrol navigation solutions lack the ability to correlate ambient temperature with risk. When the robot bypasses the obstacle, the robot cannot preferentially select one side which tends to the high-temperature target for important inspection. The perception of environmental information, avoidance of dynamic targets and the trend of high-temperature areas limit the robustness and applicability of the prior art in high-temperature dynamic industrial inspection scenes. Disclosure of Invention In order to solve the problems in the prior art, the invention provides a method for predicting obstacle avoidance and an insurance path optimization of a dynamic target of a patrol robot, which aims to realize safe and efficient patrol of the patrol robot in a dynamic high-temperature environment. The invention adopts the following technical means: The inspection robot loads the constructed static environment grid map, and adopts laser radar scanning data to match with the static environment grid map so as to perform global map positioning on the inspection robot and determine initial pose information, wherein the static environment grid map comprises fixed layout information of an inspection area; The inspection robot acquires a thermal image and a depth image of the field environment, the thermal image is segmented and binarized by adopting an entropy threshold method, and the maximum communication area is used as a high-temperature dangerous area; registering the depth image and the thermal image based on a mutual information algorithm, mapping pixel coordinates of a high-temperature target in the thermal image into the depth image through a registration relation, and obtaining three-dimensional coordinates of the high-temperature target in the depth image; Carrying out global path planning on the inspection robot by adopting a global path re-planning strategy based on predictive perception, identifying potential collision points based on a track interaction method, triggering global path dynamic re-planning, and guiding the robot to avoid dynamic targets in advance; When the global path is safely executed, the inspection robot integrates real-time temperature sensing information, quantitatively senses a high-temperature target, establishes a multi-layer temperature intensity radiation model based on a spatial gradient field, simulates a thermal diffusion effect of the high-temperature target in a real scene to dynamically optimize a local track, and guides the inspection robot to automatically face a high-temperature area. Further, in global path planning of the inspection robot, YOLOv is adopted to detect a pedestrian target, deepSORT is utilized to continuously track and smooth a track of the target across frames, a macroscopic motion displacement vector of the inspection robot is calculated based on historical track data in a section of observation window, and meanwhile, a self-adaptive correction function for vertical displacement is introduced, and a short-term position of a dynamic target is predicted based on the correction displacement vector, specifically, the method comprises the following steps: Let i be the person whose trajectory pi is a list of a series of position coordinates: Wherein, the Representing the coordinates of the pedestrian i at time t, continuously updating the position data of the pedestrian to form a series of continuous track points To capture pedestrian motion trends, define the total displacement vector within the observation window [0, t ]: The displacement vector characterizes the overall motion trend of the pedestrian i in a T frame, and in the track prediction stage, a moving path diagram of the pedestrian i is constru