Adaptive visual perception and autonomous navigation methods for dynamic industrial environments

Abstract

This invention relates to the field of autonomous navigation technology, specifically to an adaptive visual perception and autonomous navigation method for dynamic industrial environments. The method includes the following steps: acquiring a three-frame skeleton joint position sequence; determining node stability and establishing region correspondences; marking unstable node trajectories; calculating occlusion probability density; marking occluded regions; combining path segment action records and event frequency to determine stability; identifying interrupted segments; analyzing path shrinkage trends; and marking frozen state information. In this invention, by extracting the spatial displacement and direction changes of joints in personnel movements, the trend trajectory of occlusion within the navigation area is predicted, enabling pre-marking of regions in the navigation screen. By capturing continuous path segment response state changes and locking segments with continuous deviation trends, proactive identification of path instability is achieved, and abnormal path states are frozen in a timely manner, preventing the mobile platform from falling into an abnormal stagnation state due to abnormal narrowing of the local path structure.

Description

Technical Field

[0001] This invention relates to the field of autonomous navigation technology, and in particular to an adaptive visual perception and autonomous navigation method for dynamic industrial environments. Background Technology

[0002] The field of autonomous navigation technology primarily studies how mobile bodies, without external intervention or human control, can perform path planning, state estimation, obstacle avoidance control, and target guidance based on acquired environmental information. This field integrates multiple technologies, including computer vision, sensor fusion, path planning algorithms, control theory, and machine learning, aiming to improve the positioning accuracy, navigation stability, and response speed of autonomous systems in complex, dynamic, and unstructured environments. Typical applications include service robots, autonomous vehicles, intelligent manufacturing equipment, and inspection robots. Key challenges in this field include uncertainties in environmental mapping and perception, the fusion and processing of multi-source sensor data, efficient path generation mechanisms, and the integration of dynamic obstacle avoidance and real-time control strategies.

[0003] Among them, the adaptive visual perception and autonomous navigation method for dynamic industrial environments aims to construct an autonomous navigation method with visual perception capabilities that can dynamically adapt to complex changes in industrial environments. The method dynamically models the environment state by introducing visual information and combines it with autonomous navigation algorithms to achieve path replanning and behavior adjustment, supporting robots or mobile platforms to safely and stably perform movement tasks in industrial scenarios with frequent personnel activity and constantly changing obstacles. The main applications of this method include industrial production line inspection, dynamic material handling, and autonomous scheduling in flexible manufacturing scenarios, improving the flexibility and intelligence level of automation systems.

[0004] Traditional navigation methods fail to deeply identify the detailed changes in the continuous impact of personnel and environmental factors on the navigation area under continuous movement. When modeling the environment and planning navigation paths, they rely on environmental information from a single moment or short period, neglecting the specific influence of personnel movement direction, speed fluctuations, and continuous changes in spatial position on path response stability. This leads to delays and inaccuracies in judging the formation of obstructed areas and path instability trends. Traditional methods only initiate path replanning or stop movement when obstacles actually appear or the path is actually obstructed, failing to anticipate potential obstruction and path contraction risks caused by personnel movement or changes in movement. This results in frequent emergency stops and frequent path replanning, increasing wear and tear on the mobile platform and reducing task efficiency. Furthermore, the lack of accurate identification and tracking of changes in response consistency between path segments makes it impossible to proactively detect abnormal narrowing of path segments caused by continuous instability. This can easily lead to the mobile platform entering narrow spaces and being unable to recover autonomously, severely impacting the operational safety and stability of industrial field navigation systems. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and propose an adaptive visual perception and autonomous navigation method for dynamic industrial environments.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an adaptive visual perception and autonomous navigation method for dynamic industrial environments, comprising the following steps:

[0007] S1: Acquire machine vision images of industrial scenes, extract the spatial position sequence of human skeleton joints from three consecutive frames of personnel, classify joints that show sudden changes in direction and fluctuations in movement amplitude as unstable nodes, and obtain personnel movement stability information.

[0008] S2: Based on the personnel movement stability information, the corresponding area in the image used for navigation recognition in the industrial workshop is called, the occlusion probability density value is calculated, the area with occlusion trend is screened and the layer is labeled to obtain the image occlusion prediction labeling information.

[0009] S3: Obtain the image occlusion prediction annotation information, determine the response stability of the path segment in the dynamic scene based on the change in the unit time action trigger frequency between adjacent path segments, and classify the path segment into stable segments and fluctuating segments according to the change trend to obtain the path segment response consistency distribution information.

[0010] S4: Call the path segment response consistency distribution information. Based on the identified fluctuation segment, when the state of the path segment continuously deviates from the stable segment and no longer recovers to the original level, it is determined to be an execution interruption segment. Lock the path execution state and obtain the navigation path instability lock list.

[0011] As a further aspect of the present invention, the personnel movement stability information includes node fluctuation distribution, skeleton structure mapping relationship, regional movement discreteness index, position sequence overlap identifier, and spatial stability trend type. The image occlusion prediction annotation information specifically includes occlusion location tile index, node-region interaction relationship group, occlusion occurrence time period label, visual conflict direction type, and texture filling call mark. The path segment response consistency distribution information includes path segment number identifier, event trigger concentration trend, movement feedback offset mode, path segment dynamic level partition, and movement response connection relationship set. The navigation path instability lock list specifically includes path interruption segment location set, execution state deviation sequence, continuous abnormal duration interval, control command freeze label, and state inconsistency evidence group within the time period.

[0012] As a further aspect of the present invention, the step of obtaining the personnel movement stability information specifically includes:

[0013] S101: Acquire machine vision images of industrial scenes, extract the three-dimensional coordinate information of skeleton joints from each frame image, perform structural alignment of three frames of data according to node number and construct a frame sequence set, identify the continuous positions of three frames under the same number, and establish a time sequence position group of skeleton nodes.

[0014] S102: Based on the time sequence position group of the skeleton node, the displacement vector between two adjacent frames is called to calculate the angle between directions, and it is determined whether the angle meets the direction change threshold condition. The difference between the magnitudes of the displacement vectors between three frames is used to determine whether they meet the amplitude fluctuation range. Joints that meet the two conditions are selected as unstable nodes, and a joint dynamic fluctuation index table is obtained.

[0015] S103: Call the joint dynamic fluctuation index table, map the physical location of the node in the industrial scene image according to the torso, limbs and head partition category to which the node belongs in the standard skeleton structure, combine the actual positioning coordinates of the node in the navigation mapping area in the picture, establish the motion stability mapping information of the corresponding area, and generate personnel motion stability information.

[0016] As a further aspect of the present invention, the step of obtaining the image occlusion prediction annotation information specifically includes:

[0017] S201: Based on the stability information of the personnel's movements, coordinate sets are sequentially superimposed on the continuous image frames in time sequence. A set of trajectory curves under the continuous frames is established for each unstable node and mapped to the corresponding navigation area index structure in the workshop image coordinate plane to obtain the node trajectory offset mapping set.

[0018] S202: Based on the node trajectory offset mapping set, obtain the intersection segment with the image structure edge in the two-dimensional plane, filter the trajectory set whose influence value is within the preset occlusion sensitive range, calculate the occlusion probability density value of each trajectory, retain the regional trajectory whose occlusion probability density value is greater than the occlusion risk judgment threshold, and generate an occlusion trend probability distribution map.

[0019] S203: Based on the occlusion trend probability distribution map, locate the actual region boundary in the image pixel coordinate block, call the layer index information of the region to assign occlusion label value, and register the index number corresponding to the current timestamp and node trajectory in the layer cache, establish the relationship between the visible occlusion identification blocks of the region, and generate image occlusion prediction label information.

[0020] As a further aspect of the present invention, the formula for calculating the occlusion probability density value of each trajectory is specifically as follows:

[0021]

[0022] Where ρ represents the occlusion probability density value corresponding to the trajectory sequence, and n represents the number of image frames involved in the calculation. Let represent the trajectory direction vector of the unstable node at the end of the i-th frame. This represents the tangent vector of the image structure edge corresponding to the node in the i-th frame. This represents the magnitude of the node trajectory vector in the i-th frame. This represents the magnitude of the image edge vector in the i-th frame. This represents the absolute value of the dot product between the node trajectory vector and the edge vector in the i-th frame. This represents the distance from a node to an edge in the i-th frame. This represents the maximum distance between the node and the edge within three frames in the i-th frame.

[0023] As a further aspect of the present invention, the step of obtaining the path segment response consistency distribution information specifically includes:

[0024] S301: Obtain the image occlusion prediction annotation information, collect the action trigger logs of each path segment in the current industrial scene during the navigation process of the mobile platform, and classify and count the number of actions of each path segment according to the time sequence based on the action execution records of each path segment in the log within a unit time, and establish a time mapping index table to generate a set of path segment action frequencies.

[0025] S302: Based on the set of path segment action frequencies, count the number of events occurring per unit time and compare them with the path segment action frequencies, calculate and obtain the response change intensity value of each path segment, and determine the state classification of the path segment in the current dynamic scene based on whether the change intensity exceeds the response stability threshold, and obtain the path segment response state label.

[0026] S303: Based on the path segment response status label, establish a continuous number sequence for the path segments determined to be in a response fluctuation state, call the time index and spatial location information for integrated mapping, and construct spatial identifiers for stable and fluctuating segments on the path structure layer to obtain path segment response consistency distribution information.

[0027] As a further aspect of the present invention, the formula for obtaining the response change intensity value of each path segment is specifically as follows:

[0028]

[0029] Among them, f j f represents the frequency of action per unit time in the j-th path segment. j-1 This represents the frequency of action per unit time in the (j-1)th path segment. s represents the normalized mean of the frequency of actions on a path segment. j s represents the offset value of the j-th segment of personnel movement.j-1 This represents the offset value of the personnel movement in the (j-1)th segment. e represents the normalized mean of the personnel offset across the entire path segment. j Indicates the number of device status updates for segment j, e j-1 δ represents the number of device status updates in the (j-1)th segment, m represents the total number of path segments, and δ is the response change intensity value of the path segment.

[0030] As a further aspect of the present invention, the step of obtaining the navigation path instability lock list specifically includes:

[0031] S401: Call the path segment response consistency distribution information, combine the response state values ​​of adjacent path segments in time order, construct the time series of state values ​​into a path segment state evolution trajectory sequence, and mark the consistency of continuous offset direction in the sequence to generate a path state offset trend sequence.

[0032] S402: Based on the path state offset trend sequence, determine whether the state value of each path segment remains consistent with the average state deviation direction of the stable segment within the continuous sliding time window, calculate the offset interval length, filter path segments with offset length greater than the stability duration threshold, extract the number and corresponding time slice index, and establish a continuous offset index group for path segments.

[0033] S403: Call the continuous offset index group of the path segment, extract the corresponding spatial coordinate sequence information in the navigation path according to the path segment number and time slice index, mark the path segment as the execution interruption segment, and record the spatial coordinates, time nodes and status lock identifiers in the lock list mapping table to establish a navigation path instability lock list.

[0034] As a further aspect of the present invention, the method further includes the following steps:

[0035] S5: Call the navigation path instability lock list, identify whether the length change between consecutive segments shows a single-direction reduction, and make a corresponding judgment with the change trend and the record of the change in obstacle density in the scene. If the abnormal structure shrinkage characteristics are met, add a pause command mark to the path segment to obtain navigation control freeze state information.

[0036] The navigation control freeze status information includes path segment contraction trend type, environmental congestion section number, pause trigger timestamp set, path scheduling pause signal identifier, and structural constraint reinforcement label.

[0037] As a further aspect of the present invention, the step of obtaining the navigation control frozen state information specifically includes:

[0038] S501: Call the navigation path instability lock list, extract the adjacent path segment index of the path segment in the navigation path planning record according to the marked path segment coordinate sequence, calculate the Euclidean distance between the start and end nodes of the path segment and establish a length sequence to form a distance change sequence of continuous path segments and generate a path segment length change trend set.

[0039] S502: Based on the path segment length variation trend set, determine whether the length of multiple consecutive path segments shows a single-direction shrinkage trend, and mark the path segments that satisfy the decreasing relationship in sequence. Perform spatial overlap calculation between the corresponding positions of the marked sequence and the obstacle distribution area in the scene coordinate system to obtain the path segment number set that satisfies the shrinkage feature, and obtain the abnormal structure aggregated path number set.

[0040] S503: Call the abnormal structure aggregated path number set, bind the path segment execution status in the navigation control device, set the navigation signal freeze instruction flag within the corresponding time period, synchronize the status to the path execution scheduling table and generate a control record, and establish navigation control freeze status information.

[0041] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0042] In this invention, by extracting the spatial displacement and directional changes of key points in personnel movements, the stability of the movements is accurately identified, and the trend trajectory of occlusion within the navigation area is predicted. This allows for the pre-marking of areas in the navigation screen. Based on the occlusion trend and the spatial position of nodes, the stability of path segment responses is determined in real time and refined into stable and fluctuating segments. This ensures that the mobile platform clearly understands the reliability of each path segment during path selection, avoiding frequent interruptions or misoperations caused by personnel movement and changes in equipment status during navigation. By capturing changes in the response status of continuous path segments and locking down segments with a continuous deviation trend, proactive identification of path instability is achieved, and abnormal paths are frozen in a timely manner. Based on the Euclidean length trend of spatial coordinate sequence changes and the correlation with obstacle density, abnormal path contraction is accurately detected and intervened in advance, preventing the mobile platform from falling into an abnormal stagnation state due to abnormal narrowing of the local path structure. This enables the mobile platform to proactively perceive and predict risks in complex and dynamic industrial environments, significantly improving the adaptive response speed and task execution stability of the navigation system in the face of sudden environmental changes, reducing navigation risks caused by on-site personnel interference and sudden equipment conditions, and enhancing overall autonomous navigation performance. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 This is a schematic diagram of the main steps of the present invention;

[0045] Figure 2 This is a detailed schematic diagram of S1 of the present invention;

[0046] Figure 3 This is a detailed schematic diagram of S2 of the present invention;

[0047] Figure 4 This is a detailed schematic diagram of S3 of the present invention;

[0048] Figure 5 This is a detailed schematic diagram of S4 of the present invention;

[0049] Figure 6 This is a detailed schematic diagram of S5 of the present invention. Detailed Implementation

[0050] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0051] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0052] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent.

[0053] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0054] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0055] Please see Figure 1 This invention provides a technical solution: an adaptive visual perception and autonomous navigation method for dynamic industrial environments, comprising the following steps:

[0056] S1: Acquire machine vision images of industrial scenes, extract the spatial position sequence of human skeleton joints in three consecutive frames, and judge the stability of joints in continuous movements based on the displacement direction and position movement amplitude of each joint in the three frames. Joints with sudden changes in direction and fluctuations in movement amplitude are classified as unstable nodes. Based on their distribution in the skeleton structure, establish the correspondence with the current visual modeling area to obtain the stability information of human movements.

[0057] The spatial position sequence of skeleton joints corresponds to the two-dimensional or three-dimensional coordinate set output by human pose estimation tools such as OpenPose and MediaPipe. The form is the (X,Y) or (X,Y,Z) coordinates of each joint in each frame of the image. The displacement direction change can be calculated by the cosine of the angle between three frames. The change in direction can be understood as the change feature of the angle greater than π / 2 (90 degrees).

[0058] S2: Based on the stability information of personnel movements, according to the spatial location set of unstable nodes, the corresponding area in the image screen used for navigation recognition in the industrial workshop is called. The movement trend trajectory of the node in the time series is marked in the image. According to the intersection range of the node movement direction and the edge of the image structure, the occlusion probability density value is calculated, the area with occlusion trend is screened and the layer is marked to obtain the image occlusion prediction labeling information.

[0059] S3: Obtain image occlusion prediction annotation information. Based on the action trigger records of each path segment when the mobile platform performs path navigation tasks in the current industrial scenario, and based on the change in the unit time action trigger frequency between adjacent path segments, combined with the event frequency of personnel movement and equipment status update in the navigation environment during the same period, determine the response stability of the path segment in the dynamic scenario, and classify the path segment into stable segments and fluctuating segments according to the change trend to obtain the path segment response consistency distribution information.

[0060] The frequency of action triggering per unit time refers to the number of navigation control actions performed by the robot or automated navigation platform within a standard time window; the frequency of events refers to the temporal density of perceived events such as personnel detection events and changes in equipment status, which are usually recorded in the status log table.

[0061] S4: Call the path segment response consistency distribution information. Based on the identified fluctuation segment, compare the response consistency status change trend of the segment during the execution of continuous instructions. When the status of the path segment continuously deviates from the stable segment and no longer recovers to the original level, it is determined to be an execution interruption segment. Mark the spatial coordinate sequence and time node corresponding to the interruption segment, lock the path execution status, and obtain the navigation path instability lock list.

[0062] The trend of response consistency status change can be differentially processed by the unit time action frequency change of continuous path segments using a sliding window method to monitor whether the status fluctuation segment has been maintained for a period of time without recovery, thus forming the basis for interruption identification.

[0063] S5: Call the navigation path instability lock list, perform trend analysis on the Euclidean length records of the path segments in adjacent segments based on the marked path segment coordinate sequence, identify whether the length change between consecutive segments shows a single-direction reduction, and make a corresponding judgment with the change trend and the obstacle density change records in the scene. If the abnormal structure contraction characteristics are met, add a pause command mark to the path segment to obtain navigation control freeze state information.

[0064] Personnel movement stability information includes node fluctuation distribution, skeleton structure mapping relationship, regional movement discreteness index, position sequence overlap identifier, and spatial stability trend type. Image occlusion prediction annotation information specifically includes occlusion location tile index, node-region interaction relationship group, occlusion occurrence time period label, visual conflict direction type, and texture fill call mark. Path segment response consistency distribution information includes path segment number identifier, event trigger concentration trend, movement feedback offset pattern, path segment dynamic level partition, and movement response connection relationship set. Navigation path instability lock list specifically includes path interruption segment location set, execution state deviation sequence, continuous abnormal duration interval, control command freeze label, and state inconsistency evidence group within time period. Navigation control freeze state information includes path segment contraction trend type, environmental congestion segment number, pause trigger timestamp set, path scheduling pause signal identifier, and structural constraint reinforcement label.

[0065] Please see Figure 2 The specific steps for obtaining personnel movement stability information are as follows:

[0066] S101: Acquire machine vision images of industrial scenes, capture three consecutive frames of personnel in the images, extract the three-dimensional coordinate information of skeleton joints from each frame, perform structural alignment of the three frames of data according to the node number and construct a frame sequence set, identify the continuous positions of the three frames under the same number, and establish a time sequence position group of skeleton nodes.

[0067] To acquire machine vision images of industrial scenes, an RGB-D camera or a vision device with depth perception capabilities is required. The industrial site layout, including areas with personnel activity, fixed equipment, and navigation platform paths, should be uniformly captured. Three consecutive frames of image data should be acquired, and the image timestamps should be aligned. A skeleton recognition module is used to perform human pose recognition on the images, extracting the coordinate data of 17 standard skeleton key points from each frame. Based on the COCO skeleton definition, each node is numbered: head is node 0, left shoulder is node 5, right shoulder is node 6, and so on. These numbers are used for identifying and associating nodes across the three frames, constructing a 3D coordinate sequence, and recording it as a three-frame temporal skeleton data structure. For example, the coordinates of node 7 in the three frames are... For the coordinates (1.23,2.45,0.56), (1.38,2.61,0.58), and (1.65,2.74,0.59), the coordinates of the corresponding frames are recorded sequentially for the node numbered 7, and the three-frame sequence is assigned to the continuous structure array of the node numbered 7. This operation is repeated to traverse 17 nodes, constructing a complete three-frame skeleton node structure mapping. During the process, the node information extracted from each frame image is archived in units of frames. When aligning the structure, the numbering consistency is used for horizontal splicing to ensure that each node has a complete position record within the three frames. At the same time, empty nodes and missing points are filtered out, missing nodes are marked, and the set of points that failed to be constructed is excluded. Finally, the node temporal data group under the three-frame skeleton number is established, and the construction of the skeleton node temporal position group is completed.

[0068] S102: Based on the time sequence position group of skeleton nodes, according to the spatial coordinate sequence of each joint point in three frames, the displacement vector between two adjacent frames is called to calculate the direction angle, and it is determined whether the angle meets the direction change threshold condition. The difference of the displacement vector magnitude between the three frames is used to determine whether it meets the amplitude fluctuation range. Joint points that meet the two conditions are selected as unstable nodes, and the joint point dynamic fluctuation index table is obtained.

[0069] Based on the three-frame coordinate sequence of each node number recorded in the skeleton node temporal position group, a spatial direction change analysis needs to be performed on the inter-frame difference in this sequence. First, extract two consecutive vector segments for each node across the three frames, using the node number as the unit. Calculate the displacement vector directions between frame 1 and frame 2, and between frame 2 and frame 3, respectively. The direction change is evaluated using the vector angle calculation method, as shown in the formula: in The node displacement vectors from frame 1 to frame 2 and from frame 2 to frame 3 are respectively. The included angle threshold is set to 60°. If the cosine value is less than 0.5, the node is determined to have undergone a sudden change in direction. At the same time, the difference between the lengths of the two vector segments in the three frames is calculated. If the change amplitude is greater than 0.15 meters, it is marked as an amplitude fluctuation node. For example, if the node numbered 11 moves 0.12 meters and 0.31 meters between frames, the difference is 0.19 meters, which meets the set fluctuation threshold. The node is included in the fluctuation set. If a node meets both the conditions of sudden change in direction and amplitude fluctuation, it is recorded as an unstable node. Finally, all node numbers are traversed, and the index numbers of nodes that meet both conditions are written into the index table to establish a dynamic fluctuation index table for joints.

[0070] S103: Call the dynamic fluctuation index table of joint points, extract the marked unstable node numbers, map the physical location of the node in the industrial scene image according to the torso, limbs, and head partition categories to which the node belongs in the standard skeleton structure, combine the actual positioning coordinates of the node in the navigation mapping area in the picture, establish the motion stability mapping information of the corresponding area, and generate personnel motion stability information.

[0071] The unstable node numbers recorded in the dynamic fluctuation index table of joint points are called. According to the partition definition of the standard skeleton structure, the region category to which each unstable node belongs is found and it is divided into six major regions: head, torso, left upper limb, right upper limb, left lower limb, and right lower limb. After determining the region affiliation by matching the numbers, the corresponding position of the node is called in the industrial scene image coordinates, and the grid number of the navigation mapping region where the position is located in the current frame is read. Spatial mapping is performed to associate the partition category of the node in the skeleton structure with the position index structure in the industrial scene. If node number 11 belongs to the right upper limb and is located in region B7 grid in the image, the mapping pair is written into the stability mapping structure table. At the same time, the node time index frame number is marked to indicate its position in the time series. Finally, the spatial region position, time frame number, and structural classification information of all unstable nodes are organized and archived to form a multi-dimensional mapping set with node number as index, structural category as key, and image coordinate position as value, generating personnel movement stability information.

[0072] Please see Figure 3 The specific steps for obtaining image occlusion prediction annotation information are as follows:

[0073] S201: Based on the stability information of personnel movements, call the spatial position of the key points marked as unstable in the image screen, and sequentially superimpose the coordinate set in the continuous image frames in the time sequence. Establish a set of trajectory curves under the continuous frames for each unstable node, and map it to the corresponding navigation area index structure in the workshop image coordinate plane to obtain the node trajectory offset mapping set.

[0074] Based on the stability information of personnel movements, the unstable joint point number list recorded in the joint point dynamic fluctuation index table is first called. The skeleton point position information in the three frames of images corresponding to each number is retrieved. The three-dimensional coordinate set of each number in the three frames is obtained by matching the node numbers. The coordinates of the three frames are arranged in the order of the frame timestamps and stored as a set of time-series trajectories. Then, each set of trajectories is mapped to the two-dimensional image coordinate system and converted into a set of planar points corresponding to the image pixel positions. An independent two-dimensional trajectory curve set is established for each unstable node. In this process, if the node number is 12, its three-frame positions are (1.2, 2.3, 0.4), ( Given (1.3, 2.4, 0.4) and (1.6, 2.8, 0.5), the trajectory can be mapped to image coordinates (252, 326), (258, 332), and (270, 344), forming a curved path in the pixel plane. In the 3D navigation scene, a region index binding from the pixel point to the actual physical location is established, and the corresponding navigation region code is matched. For example, if the image point (258, 332) corresponds to the B3 grid region in the workshop, then the point is labeled with B3. Finally, the continuous trajectory of all unstable nodes within the time frame is generated, and the correspondence between the image and the navigation region is marked, resulting in a node trajectory offset mapping set.

[0075] S202: Based on the node trajectory offset mapping set, according to the direction vector of each trajectory, the intersection segment with the edge of the image structure in the two-dimensional plane is obtained. The intersection influence value is calculated by the angle between the direction of the node endpoint vector and the direction vector of the edge. The trajectory set with the influence value within the preset occlusion sensitive range is filtered. The occlusion probability density value of each trajectory is calculated and obtained. The region trajectory with the occlusion probability density value greater than the occlusion risk judgment threshold is retained to generate an occlusion trend probability distribution map.

[0076] The specific formula for calculating the occlusion probability density value for each trajectory is as follows:

[0077]

[0078] Where ρ represents the occlusion probability density value corresponding to the trajectory sequence, and n represents the number of image frames involved in the calculation. Let represent the trajectory direction vector of the unstable node at the end of the i-th frame. This represents the tangent vector of the image structure edge corresponding to the node in the i-th frame. This represents the magnitude of the node trajectory vector in the i-th frame. This represents the magnitude of the image edge vector in the i-th frame. This represents the absolute value of the dot product between the node trajectory vector and the edge vector in the i-th frame. This represents the distance from a node to an edge in the i-th frame. This represents the maximum distance between the node and the edge within three frames in the i-th frame;

[0079] The occlusion probability density value is used to measure the likelihood of occlusion risk occurring in the continuous trajectory of unstable human movement nodes in a dynamic industrial environment. Essentially, this value is a probabilistic representation parameter of occlusion trends in a spatiotemporally continuous image, used to predict areas in the image region where occlusion may occur. This index is essentially the mean density of occlusion behavior under trajectory projection in the image space, used to construct a layered occlusion risk map.

[0080] Calculation logic description:

[0081] The formula for the occlusion probability density value ρ is:

[0082]

[0083] The meanings of each element are as follows: The movement direction vector of the unstable node in the i-th frame; The tangent vector of the image structure edge in the i-th frame; The actual distance between the frame node and the edge; The maximum distance from the node to the edge within three frames; The dot product of the trajectory direction and the edge direction represents the degree of closeness of the included angle; weight term The higher the distance to the edge (the closer to occlusion), the larger the factor; finally, the probability density value is formed by averaging all frames.

[0084] Based on the trajectory numbers recorded in the node trajectory offset mapping set, the final segment vector coordinates of each trajectory in three consecutive frames of images are extracted, and a trajectory direction vector is constructed based on the last two points. If the position of the third frame is (268, 340) and the position of the second frame is (258, 332), then we have Calculate its modulus as After querying the image structure edge database, it was found that the endpoint of the trajectory was near the edge of the wall, and the corresponding edge direction vector was... Its mold length is dot product is After taking the absolute value, we get The cosine ratio term is then calculated as follows:

[0085]

[0086] Next, we calculate the distance term. We set the actual pixel distance from the node in the 3rd frame to the edge to be 38 pixels, and the maximum distance to the edge in the three frames to be 70 pixels. Then, after normalization... The distance term is calculated as follows:

[0087]

[0088] Multiplying the two results gives:

[0089] ρ i =0.9529·0.4571≈0.4359;

[0090] If this trajectory is the first trajectory and the total number of frames is 3, and the corresponding values ​​for the other two frames are set to 0.3121 and 0.6380, then the occlusion probability density corresponding to the complete trajectory of this node is:

[0091]

[0092] If the occlusion judgment threshold is set to 0.45, then ρ = 0.462 > 0.45, and the trajectory is judged to have an occlusion trend. Its trajectory number, image location block, and occlusion risk label are recorded and added to the occlusion trend probability distribution map. Other trajectories are fed into the calculation process in the same way and undergo filtering and labeling. Finally, the trajectories retained after filtering are used as the component data of the occlusion trend probability distribution map.

[0093] S203: Based on the occlusion trend probability distribution map, according to the trajectory position of the marked high occlusion probability area, locate the actual region boundary in the image pixel coordinate block, call the layer index information of the region to assign occlusion label value, and register the current timestamp and the index number corresponding to the node trajectory in the layer cache, establish the relationship between the visible occlusion identification blocks of the region, and generate image occlusion prediction label information;

[0094] Based on the list of high-occlusion trajectory numbers and their spatial coordinate range in the occlusion trend probability distribution map, the pixel boundaries of these occlusion trajectories in the image space are located. By matching the image region label mapping table, the region number is used for layer identification. The region index nodes in the layer management structure are read, and an occlusion risk label is added to each number. For example, if the occlusion path with node trajectory number 17 is located in image region R5 and the corresponding layer index is L3, then a risk label with number 17 is added to layer index L3. At the same time, its timestamp and trajectory number are recorded, and an occlusion patch identification record consisting of three items—timestamp, layer ID, and trajectory ID—is constructed and stored in the layer cache dataset. Finally, the spatial occlusion information is bound and labeled in the image hierarchy structure, generating image occlusion prediction label information.

[0095] Please see Figure 4 The specific steps for obtaining path segment response consistency distribution information are as follows:

[0096] S301: Obtain image occlusion prediction annotation information, collect action trigger logs of each path segment during navigation of the mobile platform in the current industrial scene, and classify and statistically analyze the number of actions of each path segment according to the time series and establish a time mapping index table to generate a set of path segment action frequencies.

[0097] After obtaining the image occlusion prediction annotation information, it is necessary to extract the action trigger data for each path segment from the navigation system action log in the industrial scenario. The action trigger data uses timestamps as the primary key and contains control signal records for each path segment during navigation, including action types such as turning, deceleration, and stopping. The data is clustered by path number to construct a unit-time action statistical sequence. For example, if path segment P3 triggers the action sequence [forward, turn, stop] within 3 seconds, its action frequency per unit time is 3 times. After collecting the action frequencies of all path segments, a data table structure sorted by time order is established. Simultaneously, the time index corresponding to each data segment is obtained for subsequent state evolution analysis. The constructed path segment action frequency set is f1, f2, ..., f m ], where m is the total number of path segments. The action frequencies of P1 to P5 are set to 5, 4, 6, 3, 2 respectively. Then this set is [5, 4, 6, 3, 2], which is used to support subsequent trend analysis and response status judgment, and finally generate the path segment action frequency set.

[0098] S302: Based on the set of path segment action frequencies, select the action frequency data of adjacent path segments, call the continuous position offset and equipment status update sequence in the personnel movement record, count the number of events occurring per unit time and compare them with the path segment action frequency, calculate and obtain the response change intensity value of each path segment, and determine the state classification of the path segment in the current dynamic scene based on whether the change intensity exceeds the response stability threshold, and obtain the path segment response status label.

[0099] The specific formula for calculating the response change intensity value of each path segment is as follows:

[0100]

[0101] Among them, f j f represents the frequency of action per unit time in the j-th path segment. j-1 This represents the frequency of action per unit time in the (j-1)th path segment. s represents the normalized mean of the frequency of actions on a path segment. j s represents the offset value of the j-th segment of personnel movement. j-1 This represents the offset value of the personnel movement in the (j-1)th segment. e represents the normalized mean of the personnel offset across the entire path segment. j Indicates the number of device status updates for segment j, e j-1 This represents the number of device status updates in the (j-1)th segment, m represents the total number of path segments, and δ is the response change intensity value of the path segment.

[0102] The response change intensity value is a time-series variability metric used in dynamic path analysis to measure the overall fluctuation of action feedback, personnel offset behavior, and equipment status changes in different path segments per unit time. This value is a key intermediate quantity for assessing the stability of path segments in dynamic industrial scenarios, used to divide path segments into stable and fluctuating segments. Its essence is a fusion difference intensity assessment of the changing trends of multi-source event characteristics, including changes in action triggering frequency; changes in personnel offset; and changes in equipment response status.

[0103] Calculation logic description:

[0104] The formula for the intensity value δ of the response change is:

[0105]

[0106] The definitions of each item are as follows: f j : The trigger frequency of the action per unit time in segment j; s j : The personnel offset value of segment j; e j : Number of device status updates in segment j; The model uses the normalized average of the corresponding indicators; each sub-item measures the severity of three core behaviors—behavioral frequency fluctuation, human movement stability fluctuation, and equipment status response fluctuation—between adjacent path segments; all items are normalized and processed using a differencing method, which acts as a high-pass filter to amplify "abrupt changes," effectively capturing path instability signals. The model integrates three general feature indicators (behavioral frequency, displacement, and equipment events) to form a unified intensity expression, making it a reasonable and commonly used fusion metric in industrial dynamic path response models.

[0107] Continuous path segments are selected based on the set of path segment action frequencies. The action frequency values ​​of adjacent path segments are extracted as frequency change items. Simultaneously, the coordinate displacement difference in personnel movement records is obtained as an offset item, and the statistical values ​​of equipment alarms, start-ups, and shutdowns in the equipment status log are used as status update items. These three types of data are then combined in an occlusion scenario for joint analysis. For example, the action frequency from segment 2 to segment 3 is 4 to 6, corresponding to a frequency difference of 2. The personnel offset is from (1.2, 0.9) to (1.9, 1.0), the Euclidean distance is 0.707 meters, and there are 2 equipment status change records. A normalized mean of the path segment action frequency is set. Personnel offset normalized mean Substitute into the formula to calculate:

[0108]

[0109] The calculation yields:

[0110] Finally, δ≈0.3846+0.3847+0.6667=1.436. If the response stability threshold is set to 1.1, the state of this path segment is classified as a fluctuating segment, and it is marked as an abnormal trajectory segment. A path segment response state label is constructed.

[0111] S303: Based on the path segment response status label, establish a continuous number sequence for path segments that are determined to be in a response fluctuation state, call the time index and spatial location information for integration and mapping, and construct spatial identifiers for stable and fluctuating segments on the path structure layer to obtain path segment response consistency distribution information.

[0112] Based on the list of path segments marked as fluctuating in the path segment response status labels, they are consecutively numbered according to timestamp and path segment number. The start and end coordinates of the path segment corresponding to the number on the navigation path are extracted to establish a path segment position dataset in the scene coordinate system. Spatial structure layer mapping is performed according to the number order. Stable segments and fluctuating segments are written into the layer mapping table with different label numbers. For example, stable segments are labeled as S and fluctuating segments are labeled as W. If path segments P1-P2 are stable and P3-P4 are fluctuating, then the label sequence [S,S,W,W,S] is generated and written into the corresponding coordinate structure layer to realize layer state partition mapping, and finally obtain the path segment response consistency distribution information.

[0113] Please see Figure 5 The specific steps for obtaining the navigation path instability lock list are as follows:

[0114] S401: Call the path segment response consistency distribution information, combine the response state values ​​of adjacent path segments in chronological order according to the path segment number marked as fluctuating state, construct the time series of state values ​​into a path segment state evolution trajectory sequence, and mark the consistency of continuous offset direction in the sequence to generate a path state offset trend sequence.

[0115] The path segment numbers marked as fluctuating states in the path segment response consistency distribution information are called. First, a number sequence is established and each path segment is sorted in ascending order by timestamp. The response state value of each path segment in the time evolution is extracted to construct the state time series of continuous path segments. For example, if the path segment numbers are P3, P4, and P5 and their state values ​​are 0.6, 0.65, and 0.7 respectively, then the state evolution sequence [0.6, 0.65, 0.7] can be constructed. In the sequence, the offset direction consistency detection is performed, that is, it is judged whether the sign of the difference between each two adjacent state values ​​is the same. If the signs are consistent, it is considered that the continuous offset direction is consistent. For example, in the above sequence, each time the state value increases, the direction is consistent and is recorded as positive offset. If 0.68, 0.64, and 0.60 appear later, the direction changes continuously and is negative offset, which is marked as reverse offset state. The continuous offset segments of the entire state time series are marked according to this logic and the start and end indexes of the segments are extracted. Finally, the path numbers and time slice information of the continuous segments with consistent offset direction are written into the sequence structure to generate the path state offset trend sequence.

[0116] S402: Based on the path state offset trend sequence, determine whether the state value of each path segment remains consistent with the average state deviation direction of the stable segment within the continuous sliding time window, calculate the offset interval length, filter path segments with offset length greater than the stability duration threshold, extract the number and corresponding time slice index, and establish a continuous offset index group for path segments.

[0117] Based on the path state offset trend sequence, the fluctuation range of each path segment state within the sliding window is judged. The sliding window width is set to 3. The difference between the current state of each path segment and the average state of the stable segment within the sliding window is checked for consistency. If the direction is consistent, it is judged that the offset direction is maintained. The length of the consecutive interval with consistent direction is used as the offset interval length. The path segment state value is set to [0.42, 0.48, 0.55, 0.61, 0.64], and the average value of the stable segment is 0.40. Then the difference between each state value and the average value is [+0.02, +0.08, +0.15, +0.21, +0.24], all in positive direction. The continuous offset length is 5. The stability duration threshold is set to 3. Since 5 is greater than the threshold, the path segment sequence is judged to be a continuously offset state path segment. Its number, start frame index, end frame index, etc. are recorded in the structure list to construct the path segment continuous offset index group.

[0118] S403: Call the continuous offset index group of the path segment, extract the corresponding spatial coordinate sequence information in the navigation path according to the path segment number and time slice index, mark the path segment as the execution interruption segment, and record the spatial coordinates, time nodes and status lock identifiers in the lock list mapping table to establish the navigation path instability lock list.

[0119] The path segment number and time index extracted from the continuous offset index group are called. Combined with the navigation path structure information table, the spatial coordinates of the start and end points of the path segment are retrieved. The segments are classified and recorded as continuous unstable segments according to their numbers. The task scheduling records corresponding to the time slices in the control log are read, and the task execution number, path segment execution identifier, operation node and other contents of the segment are extracted and integrated synchronously. The path segment is marked as an execution interruption segment. At the same time, a lock status identifier is established and bound to the path scheduling module. Finally, the four types of data, namely path segment number, coordinate information, time slice index, and lock status identifier, are summarized and written into the lock information structure table to establish a navigation path instability lock list.

[0120] Please see Figure 6 The specific steps for obtaining navigation control freeze status information are as follows:

[0121] S501: Call the navigation path instability lock list, extract the index of the adjacent path segment in the navigation path planning record based on the marked path segment coordinate sequence, calculate the Euclidean distance between the start and end nodes of the path segment and establish the length sequence, form the distance change sequence of continuous path segments, and generate the path segment length change trend set.

[0122] The coordinate sequence of the marked path segments in the navigation path instability lock list is called first. The start and end coordinates of each path segment are extracted, and a three-dimensional coordinate representation is constructed (x... s ,y s ,z s ), (x e ,y e ,z e The algorithm retrieves the adjacent path segment numbers from the path planning record file, obtains the corresponding spatial location point pairs through the path segment index, and combines the start and end points of each path segment as the distance calculation input. It then uses the Euclidean formula to calculate the length value. For example, if the start point of path segment P1 is (2.0, 1.0, 0.0) and the end point is (3.0, 2.0, 0.0), then its length is... This process iterates through all path segments and constructs a set of length sequences. For example, if the lengths of consecutive segments P1, P2, and P3 are 1.414, 1.322, and 1.220, they are recorded as [1.414, 1.322, 1.220]. This process ultimately generates a set of path segment length variation trends.

[0123] S502: Based on the path segment length variation trend set, call the length difference of adjacent path segments to determine whether the length of multiple consecutive path segments shows a single-direction shrinkage trend, and mark the path segments that satisfy the decreasing relationship. Call the obstacle density change record, and calculate the spatial overlap between the marked sequence and the obstacle distribution area in the scene coordinate system to obtain the path segment number set that satisfies the shrinkage feature, and obtain the abnormal structure aggregated path number set.

[0124] Based on the length values ​​of every two adjacent path segments in the path segment length variation trend set, calculate their difference sequence. For example, the difference between P1 and P2 is 1.414 + 1.322 = 0.092, and the difference between P2 and P3 is 1.322 + 1.220 = 0.102. Determine whether the sequence is a continuous decreasing relationship, that is, determine whether all differences are positive and whether they continuously exceed the set judgment threshold. The continuous decreasing threshold is set to 2. If two consecutive differences are greater than 0, the decreasing trend is considered valid. Further call the obstacle density change record file to extract the obstacle number density data corresponding to the spatial range of the path segment in the scene coordinates. For example, if the average number of obstacles in the area covered by path segment P2 is 15 per square meter, and the density threshold is 12, then the area meets the obstacle density judgment standard. Perform overlap judgment, that is, cross-calculate the path segment coordinate range and the obstacle distribution grid. If the overlap area ratio exceeds 50%, the path segment is judged as an abnormal shrinkage area. Extract its path segment number and add it to the abnormal structure aggregate path number set.

[0125] S503: Call the abnormal structure aggregated path number set, bind the path segment execution status in the navigation control device, set the navigation signal freeze instruction flag within the corresponding time period, synchronize the status to the path execution scheduling table and generate control records, and establish navigation control freeze status information;

[0126] The system calls the list of numbers extracted from the aggregated path numbers of the abnormal structure, accesses the execution control dictionary in the navigation control module one by one, sets the execution status field to "frozen" for each path segment, and adds a freeze time range field. This range is taken from the timestamp index in the instability lock list. At the same time, a freeze task record is added to the path scheduling table, which includes metadata such as path segment number, freeze time, freeze reason identifier, and instruction lock number, and is associated with the scheduling flowchart node. Once the path segment is reached during task execution, it enters the freeze waiting state. The freeze signal is recorded as FZ status in the control record. The binding relationship between this status and the path segment and the timestamp information are recorded, and finally, the navigation control freeze status information is established.

[0127] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0128] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0129] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0130] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0131] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0132] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0133] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0134] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0135] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0136] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for adaptive vision perception and autonomous navigation for dynamic industrial environments, characterized in that, The method comprises the following steps: S1: obtaining a machine vision image of an industrial scene, extracting a human body skeleton joint space position sequence of three consecutive frames of a person, classifying the joints with direction mutations and movement amplitude fluctuations as unstable nodes, and obtaining person action stability information; S2: based on the person action stability information, corresponding calling of a corresponding region in an image screen for navigation recognition in an industrial workshop, calculation of an occlusion probability density value, screening of a region with an occlusion trend and layer labeling, and obtaining image occlusion pre-judgment labeling information; The acquisition step of the image occlusion pre-judgment labeling information is specifically: S201: based on the person action stability information, sequentially superimposing coordinate sets in consecutive image frames in time sequence order, establishing a trajectory curve set under consecutive frames for each unstable node, and mapping to a corresponding navigation region index structure in a workshop image coordinate plane to obtain a node trajectory offset mapping set; S202: based on the node trajectory offset mapping set, obtaining a cross section with an image structure edge in a two-dimensional plane, screening a trajectory set with an influence value in a preset occlusion sensitive interval, calculating an occlusion probability density value of each trajectory, retaining a region trajectory with an occlusion probability density value greater than an occlusion risk judgment threshold, and generating an occlusion trend probability distribution map; S203: based on the occlusion trend probability distribution map, positioning an actual region boundary in an image pixel coordinate block, calling layer index information of the region to assign an occlusion label, registering a current timestamp and a node trajectory corresponding index number in a layer cache, establishing a visible occlusion identification block relationship of the region, and generating image occlusion pre-judgment labeling information; S3: obtaining the image occlusion pre-judgment labeling information, judging a response stability degree of a path segment in a dynamic scene according to a unit time action trigger frequency change between adjacent path segments, and classifying the path segment into a stable section and a fluctuation section according to a change trend to obtain path segment response consistency distribution information; S4: calling the path segment response consistency distribution information, according to the identified fluctuation section, when the state of the path segment continuously deviates from the stable section and no longer recovers to the original level, determining it as an execution interruption section, locking the path execution state, and obtaining a navigation path instability locking list.

2. The adaptive vision perception and autonomous navigation method for dynamic industrial environment of claim 1, wherein, The person action stability information includes node fluctuation distribution, skeleton structure mapping relationship, region action dispersion index, position sequence overlap identifier, and space stability trend type, the image occlusion pre-judgment labeling information is specifically an occlusion position block index, a node-region interaction relationship group, an occlusion occurrence time period label, a visual conflict direction type, and a texture filling calling mark, the path segment response consistency distribution information includes a path segment number identifier, a trend in an event trigger set, an action feedback offset mode, a path segment dynamic level partition, and an action response connection relationship set, and the navigation path instability locking list is specifically a path interruption section position set, an execution state deviation sequence, a continuous abnormal duration interval, a control instruction freezing label, and a state inconsistency evidence group in a time period.

3. The adaptive vision perception and autonomous navigation method for dynamic industrial environment of claim 2, wherein, The acquisition step of the person action stability information is specifically: S101: Obtain machine vision images of an industrial scene, and extract three-dimensional coordinate information of skeleton joint nodes from each frame of image, perform structure alignment on three frames of data according to node numbers, and construct a frame sequence set, identify three consecutive positions under the same number, and establish a skeleton node time sequence position group; S102: Based on the skeleton node time sequence position group, call the direction included angle of the displacement vector between adjacent two frames, judge whether the angle meets the direction mutation threshold condition, and judge whether the displacement vector length between three frames meets the amplitude fluctuation range, screen the joint nodes meeting the two conditions as unstable nodes, and obtain a joint node dynamic fluctuation index table; S103: Call the joint node dynamic fluctuation index table, map the physical area position of the node in the industrial scene image according to the trunk, limb and head part area category to which the node belongs in the standard skeleton structure, combine the actual positioning coordinates of the navigation mapping area where the node is located in the picture, establish the action stability mapping information of the corresponding area, and generate personnel action stability information.

4. The adaptive vision perception and autonomous navigation method for dynamic industrial environment of claim 1, wherein, The formula for calculating the occlusion probability density value of each trajectory is: ； in, This represents the occlusion probability density value corresponding to the trajectory sequence. This indicates the number of image frames involved in the calculation. Indicates the first The trajectory direction vector of the unstable node at the end of the frame. Indicates the first The tangent vectors of the edges of the image structure corresponding to the nodes in the frame. Indicates the first The magnitude of the node trajectory vector in the frame. Indicates the first The magnitude of the edge vector of the image in the frame. Indicates the first The absolute value of the dot product between the node trajectory vector and the edge vector in the frame. Indicates the first The distance value from a node in a frame to the edge. Indicates the first The maximum distance between the node and the edge within three frames.

5. The adaptive vision perception and autonomous navigation method for dynamic industrial environment of claim 4, wherein, The acquisition step of the path segment response consistency distribution information is: S301: Obtain the image occlusion pre-judgment labeling information, collect the action trigger log of each path segment of the moving platform in the navigation process in the current industrial scene, classify and statistically establish a time mapping index table according to the action execution record corresponding to each path segment in the log within a unit time, and generate a path segment action frequency set; S302: Based on the path segment action frequency set, respectively, the number of events occurring within a unit time is counted and compared with the path segment action frequency, the response change intensity value of each path segment is calculated, and the state classification of the path segment in the current dynamic scene is judged according to whether the change intensity exceeds the response stability threshold, and the path segment response state label is obtained; S303: According to the path segment response state label, a continuous number sequence is established for the path segment in the response fluctuation state, the time index and the spatial position information are integrated and mapped, and the spatial identification of the stable section and the fluctuation section on the path structure layer is constructed, and the path segment response consistency distribution information is obtained.

6. The adaptive visual perception and autonomous navigation method for dynamic industrial environment according to claim 5, wherein the formula for calculating the response change intensity value of each path segment is: ； wherein, represents the number of actions per time unit for the path segment, represents the number of actions per time unit for the path segment, represents the number of actions per time unit for the path segment, represents the number of actions per time unit for the path segment, represents the normalized mean of the path segment action frequency, represents the number of actions per time unit for the path segment, represents the number of actions per time unit for the path segment, represents the number of actions per time unit for the path segment, represents the number of actions per time unit for the path segment, represents the normalized mean of the path segment action frequency, represents the number of actions per time unit for the path segment, represents the number of actions per time unit for the path segment, represents the number of actions per time unit for the path segment, represents the number of actions per time unit for the path segment, represents the total number of path segments, is a response change intensity value for the path segment.

7. The adaptive vision perception and autonomous navigation method for dynamic industrial environment of claim 6, wherein, The acquisition step of the navigation path instability locking list is: S401: Call the path segment response consistency distribution information, combine the response state values of adjacent path segments in time sequence, construct the path segment state evolution trajectory sequence from the time sequence of the state values, and label the consistency of the continuous offset direction in the sequence, and generate a path state offset trend sequence; S402: According to the path state offset trend sequence, it is judged whether the state value of each path segment remains consistent with the average state deviation direction of the stable segment in the continuous sliding time window, the offset interval length is calculated, the path segments with an offset length greater than the stability duration threshold are screened, the number and corresponding time slice index are extracted, and the path segment continuous offset index group is established; S403: The path segment continuous offset index group is called, the corresponding spatial coordinate sequence information in the navigation path is extracted according to the path segment number and the time slice index, the path segment is marked as an execution interruption segment, and the spatial coordinates, the time node and the state locking identification are uniformly recorded in the locking list mapping table, and the navigation path instability locking list is established.

8. The adaptive vision perception and autonomous navigation method for dynamic industrial environment of claim 7, wherein, The method further comprises the following steps: S5: The navigation path instability locking list is called, it is judged whether the length change between the continuous segments presents a single direction reduction, and the change trend is correspondingly judged with the obstacle density change record in the scene, if the abnormal structure contraction feature is met, a pause instruction identification is added to the path segment, and navigation control freezing state information is obtained; The navigation control freezing state information comprises a path segment contraction trend type, an environment congestion segment number, a pause trigger timestamp set, a path scheduling pause signal identification and a structure constraint strengthening label.

9. The adaptive vision perception and autonomous navigation method for dynamic industrial environment of claim 8, wherein, The acquisition step of the navigation control freezing state information is specifically: S501: The navigation path instability locking list is called, the adjacent path segment index of the path segment in the navigation path planning record is extracted according to the marked path segment coordinate sequence, the Euclidean distance between the start and end nodes of the path segment is calculated and a length sequence is established, the distance change sequence of the continuous path segment is constructed, and a path segment length variation trend set is generated; S502: According to the path segment length variation trend set, it is judged whether the length of the continuous multiple path segments presents a single direction reduction trend, the path segments satisfying the decreasing relationship are sequentially marked, the spatial overlap calculation between the marked sequence corresponding position in the scene coordinate system and the obstacle distribution segment is performed, the path segment number set satisfying the contraction feature is obtained, and an abnormal structure aggregation path number set is obtained; S503: The abnormal structure aggregation path number set is called, the path segment execution state is bound in the navigation control device, the navigation signal freezing instruction identification in the corresponding time period is set, the state is synchronized to the path execution scheduling table and the control record is generated, and the navigation control freezing state information is established.

Images