Medical logistics robot route planning method and system based on adaptive feedback

Abstract

The present invention relates to the field of path planning technology and invents a medical logistics robot route planning method based on adaptive feedback, comprising: topologically reconstructing and path planning a hospital static route map based on real-time dynamic logistics information to obtain a primary logistics route sequence; selecting primary logistics routes in the primary logistics route sequence one by one, performing obstacle recognition and dynamic object tracking on a real-scene image sequence to obtain a real-time obstacle image sequence; performing three-dimensional mapping and motion modeling on the real-time obstacle image sequence based on a real-time point cloud sequence to obtain an obstacle motion path; performing heuristic sampling and iterative path search on the primary logistics routes based on the obstacle motion path to obtain a standard logistics route, and performing global evaluation and optimization on the primary logistics route sequence to obtain a standard logistics route sequence. The present invention also proposes a medical logistics robot route planning system based on adaptive feedback. The present invention can improve the flexibility of route planning.

Description

Technical Field

[0001] The present invention relates to the field of path planning technology, and in particular to a medical logistics robot route planning method and system based on adaptive feedback. Background Art

[0002] Medical logistics robots are able to accurately carry a variety of medical supplies such as medicines, medical devices, and test samples. With their advanced mechanical structure and control system, medical logistics robots can stably carry items of different weights and sizes, ensuring that the items are intact during transportation.

[0003] In order to achieve efficient, accurate and safe logistics distribution, medical logistics robots need to plan routes. Existing sales analysis methods usually rely on static maps for path planning. In actual applications, the hospital environment is relatively complex, with a large flow of people and a large number of moving obstacles. The existing route planning methods that rely on static maps lack the ability to adapt to dynamically changing environments, which may lead to low flexibility in route planning. Summary of the Invention

[0004] The present invention provides a medical logistics robot route planning method and system based on adaptive feedback, the main purpose of which is to solve the problem of low flexibility during route planning.

[0005] To achieve the above objectives, the present invention provides a medical logistics robot route planning method based on adaptive feedback, comprising:

[0006] Obtaining a hospital static route map and real-time dynamic logistics information, and performing topological reconstruction and path planning on the hospital static route map based on the real-time dynamic logistics information to obtain a primary logistics route sequence;

[0007] Selecting primary logistics routes from the primary logistics route sequence one by one, performing path navigation and real-scene sensing monitoring according to the primary logistics routes, and obtaining a real-scene cloud sequence and a real-scene image sequence;

[0008] Performing obstacle recognition and dynamic object tracking on the real scene image sequence to obtain a real-time obstacle image sequence;

[0009] performing three-dimensional mapping and motion modeling on the real-time obstacle image sequence according to the real-time point cloud sequence to obtain an obstacle motion path, wherein the performing three-dimensional mapping and motion modeling on the real-time obstacle image sequence according to the real-time point cloud sequence to obtain the obstacle motion path includes: extracting a point cloud feature point sequence from the real-time point cloud sequence and extracting an obstacle feature point sequence from the real-time obstacle image sequence; performing feature point matching and coordinate alignment on the obstacle feature point sequence according to the point cloud feature point sequence to obtain a coordinate mapping relationship; mapping the real-time obstacle image sequence to the real-time point cloud sequence according to the coordinate mapping relationship to obtain a mapped point cloud sequence; performing point cloud denoising on the mapped point cloud sequence to obtain an obstacle point cloud sequence; performing three-dimensional reconstruction on the obstacle point cloud sequence to obtain an obstacle model sequence; performing motion estimation on the obstacle model sequence to obtain an obstacle motion model; and performing path mapping and path smoothing on the obstacle motion model to obtain an obstacle motion path;

[0010] The primary logistics route is sampled heuristically and searched iteratively according to the obstacle movement path to obtain a standard logistics route, and the primary logistics route sequence is globally evaluated and optimized according to the real-time dynamic logistics information and the standard logistics route to obtain a standard logistics route sequence.

[0011] Optionally, the topology reconstruction and path planning of the hospital static route map are performed according to the real-time dynamic logistics information to obtain a primary logistics route sequence, including:

[0012] Extracting a static node set and a static route set from the hospital static route map respectively;

[0013] Extracting a route attribute set corresponding to the static route set from the real-time dynamic logistics information;

[0014] Performing route feature extraction and route weight mapping on the route attribute set to obtain a route weight set;

[0015] Performing topology reconstruction based on the static node set and the static route set to obtain a primary route topology graph;

[0016] Performing path weighting on the primary route topology map using the route weight set to obtain a standard route topology map;

[0017] Extracting a logistics starting point and a logistics end point from the real-time dynamic logistics information;

[0018] Path planning is performed on the standard route topology map according to the logistics starting point and the logistics end point to obtain a primary logistics route sequence.

[0019] Optionally, performing obstacle recognition and dynamic object tracking on the real-scene image sequence to obtain a real-time obstacle image sequence includes:

[0020] Performing edge detection and road fitting on the real scene image sequence to obtain real scene road edges;

[0021] Performing primary obstacle detection on the real scene image sequence to obtain a primary obstacle detection frame sequence;

[0022] Performing motion analysis on the primary obstacle detection frame sequence to obtain an analyzed obstacle detection frame sequence;

[0023] performing data association and updating on the primary obstacle detection frame sequence and the analysis obstacle detection frame sequence to obtain an associated obstacle detection frame sequence;

[0024] Performing obstacle screening on the associated obstacle detection frame sequence according to the real-scene road edge to obtain a real-time obstacle detection frame sequence;

[0025] A real-time obstacle image sequence is extracted from the real-scene image sequence using the real-time obstacle detection frame sequence.

[0026] Optionally, performing obstacle detection on the real scene image sequence to obtain a primary obstacle detection frame sequence includes:

[0027] Selecting real scene images in the real scene image sequence one by one as target real scene images, performing adaptive residual convolution on the target real scene images, and obtaining residual real scene features;

[0028] Performing layer-by-layer multi-scale downsampling on the residual real scene features to obtain a multi-scale real scene feature set;

[0029] Performing feature slicing and local attention encoding on the multi-scale real scene feature set to obtain a local real scene feature set;

[0030] Performing global attention encoding on the local real scene feature set to obtain a global real scene feature set;

[0031] Performing multi-level feedforward activation and residual connection on the global real scene feature set to obtain an enhanced real scene feature set;

[0032] Performing layer-by-layer upsampling and obstacle feature activation on the enhanced real scene feature set to obtain a primary obstacle detection frame;

[0033] The activated obstacle detection frames corresponding to all target real scene images in the real scene image sequence are collected into a primary obstacle detection frame sequence.

[0034] Optionally, performing layer-by-layer upsampling and obstacle feature activation on the augmented reality feature set to obtain a primary obstacle detection frame includes:

[0035] Sorting the enhanced real scene feature set in ascending order of feature size to obtain an enhanced real scene feature sequence;

[0036] Selecting the first enhanced real scene feature in the enhanced real scene feature sequence as the target first real scene feature, and selecting the enhanced real scene feature following the first real scene feature in the enhanced real scene feature sequence as the second real scene feature;

[0037] Using the characteristic size of the second real scene feature as the target characteristic size;

[0038] Upsampling the first real scene feature according to the target feature size to obtain a target upsampled real scene feature;

[0039] Performing feature fusion on the target upsampled real scene feature and the second real scene feature to obtain a target fusion feature;

[0040] Determining whether the second real scene feature is the last enhanced real scene feature in the enhanced real scene feature sequence;

[0041] If not, then using the target fusion feature to update the target first real scene feature, taking the enhanced real scene feature located after the second real scene feature in the enhanced real scene feature sequence as the second real scene feature, and returning to the step of taking the feature size of the second real scene feature as the target feature size;

[0042] If so, feature decoding and feature activation are performed on the target fusion feature to obtain a primary obstacle detection frame.

[0043] Optionally, the performing data association updating on the primary obstacle detection frame sequence and the analyzed obstacle detection frame sequence to obtain a real-time obstacle detection frame sequence includes:

[0044] selecting primary obstacle detection frames in the primary obstacle detection frame sequence one by one as target primary detection frames, and using analysis obstacle detection frames corresponding to the target primary detection frames in the analysis obstacle detection frame sequence as target analysis detection frames;

[0045] Extracting a primary obstacle image from the real scene image sequence according to the target primary detection frame, and extracting an analysis obstacle image from the real scene image sequence according to the target analysis detection frame;

[0046] extracting primary obstacle features from the primary obstacle image and extracting analytical obstacle features from the analytical obstacle image;

[0047] Calculating a detection frame distance between the target primary detection frame and the target analysis detection frame;

[0048] Calculating a characteristic distance between the primary obstacle feature and the analysis obstacle feature;

[0049] The position of the target analysis detection frame is updated according to the detection frame distance and the feature distance to obtain a real-time obstacle detection frame;

[0050] The real-time obstacle detection frames corresponding to all target analysis detection frames in the analysis obstacle detection frame sequence are aggregated into a real-time obstacle detection frame sequence.

[0051] Optionally, performing heuristic sampling and iterative path search on the primary logistics route according to the obstacle movement path to obtain a standard logistics route includes:

[0052] Obtaining a route map of the primary logistics route;

[0053] Dividing the route map into grids according to a preset robot size to obtain a logistics map grid;

[0054] Extracting a route end point and a route starting point from the primary logistics route;

[0055] Marking the points of the logistics map grid according to the route end point and the route starting point to obtain a marked map grid;

[0056] De-weighting the marked map grid according to the route endpoint to obtain a primary sampling grid;

[0057] Performing weighted sampling on the primary sampling grid according to the obstacle motion path to obtain a standard sampling grid;

[0058] An iterative path search is performed on the standard sampling grid according to the obstacle movement path to obtain a standard logistics route.

[0059] Optionally, performing an iterative path search on the standard sampling grid according to the obstacle movement path to obtain a standard logistics route includes:

[0060] Initializing a particle group, selecting particles in the particle group one by one as target particles, and using the starting point of the route in the standard sampling grid as the initial position of the target particle;

[0061] Calculating a grid movement interval based on a preset robot movement speed and the standard sampling grid;

[0062] Sampling the obstacle motion path frame by frame according to the grid movement interval to obtain a target frame obstacle grid;

[0063] Performing safety area screening on the standard sampling grid according to the target frame obstacle grid to obtain a safe sampling grid;

[0064] updating the position of the target particle in the safety sampling grid one by one to obtain a target update position, and generating a moving route of the target particle according to all target update positions and the initial position;

[0065] Determining whether the target update position is located at a route end point in the standard sampling grid graph;

[0066] If not, returning to the step of sampling the obstacle motion path frame by frame according to the grid movement interval to obtain the target frame obstacle grid;

[0067] If yes, the moving route is used as the initial search route of the target particle;

[0068] Aggregating the initial search routes of all target particles in the particle group into an initial search route set;

[0069] The initial search route set is screened for the shortest route to obtain a standard logistics route.

[0070] Optionally, performing global evaluation and optimization on the primary logistics route sequence according to the real-time dynamic logistics information and the standard logistics route to obtain a standard logistics route sequence includes:

[0071] Determining whether the primary logistics route corresponding to the standard logistics route is the last primary logistics route in the primary logistics route sequence;

[0072] If not, then perform a time analysis on the standard logistics route to obtain the standard logistics time;

[0073] The real-time dynamic logistics information is updated according to the standard logistics duration to obtain standard dynamic logistics information;

[0074] Perform route overlay on the hospital static route map according to the standard logistics route to obtain a covered static route map;

[0075] Performing topology reconstruction and path planning on the covered static route map according to the standard dynamic logistics information to obtain an updated logistics route sequence;

[0076] Using the updated logistics route sequence to update the primary logistics route sequence, and returning to the step of selecting primary logistics routes in the primary logistics route sequence one by one;

[0077] If so, the standard logistics routes corresponding to all primary logistics routes in the primary logistics route sequence are aggregated into a standard logistics route sequence.

[0078] In order to solve the above problems, the present invention also provides a medical logistics robot route planning system based on adaptive feedback, the system comprising:

[0079] A topology planning module is used to obtain a hospital static route map and real-time dynamic logistics information, and to perform topology reconstruction and path planning on the hospital static route map based on the real-time dynamic logistics information to obtain a primary logistics route sequence;

[0080] A real-scene monitoring module is used to select primary logistics routes from the primary logistics route sequence one by one, perform path navigation and real-scene sensing monitoring according to the primary logistics routes, and obtain a real-scene point cloud sequence and a real-scene image sequence;

[0081] An obstacle tracking module is used to perform obstacle recognition and dynamic object tracking on the real scene image sequence to obtain a real-time obstacle image sequence;

[0082] a motion modeling module, configured to perform three-dimensional mapping and motion modeling on the real-time obstacle image sequence based on the real-time point cloud sequence to obtain an obstacle motion path, wherein the performing three-dimensional mapping and motion modeling on the real-time obstacle image sequence based on the real-time point cloud sequence to obtain the obstacle motion path comprises: extracting a point cloud feature point sequence from the real-time point cloud sequence and extracting an obstacle feature point sequence from the real-time obstacle image sequence; performing feature point matching and coordinate alignment on the obstacle feature point sequence based on the point cloud feature point sequence to obtain a coordinate mapping relationship; mapping the real-time obstacle image sequence to the real-time point cloud sequence based on the coordinate mapping relationship to obtain a mapped point cloud sequence; performing point cloud denoising on the mapped point cloud sequence to obtain an obstacle point cloud sequence; performing three-dimensional reconstruction on the obstacle point cloud sequence to obtain an obstacle model sequence; performing motion estimation on the obstacle model sequence to obtain an obstacle motion model; and performing path mapping and path smoothing on the obstacle motion model to obtain an obstacle motion path;

[0083] A global optimization module is used to perform heuristic sampling and iterative path search on the primary logistics route according to the obstacle movement path to obtain a standard logistics route, and to perform global evaluation and optimization on the primary logistics route sequence according to the real-time dynamic logistics information and the standard logistics route to obtain a standard logistics route sequence.

[0084] The present invention obtains a primary logistics route sequence by topologically reconstructing and path planning the hospital static route map according to the real-time dynamic logistics information. It can realize global weighted path planning in combination with the complex road environment in the hospital, thereby determining the primary robot driving path and improving the flexibility of path planning. By performing path navigation and real-scene sensing monitoring according to the primary logistics route, a real-scene point cloud sequence and a real-scene image sequence are obtained, and point cloud data and real-scene images of the current driving path can be obtained, thereby realizing multi-dimensional detailed perception of the surrounding environment, obtaining more accurate environmental information, and improving the accuracy of adaptive feedback and path planning. By performing obstacle recognition and dynamic object tracking on the real-scene image sequence, obstacles that may hinder the medical logistics robot in the real-scene area can be effectively identified, and the motion tracking of the obstacles can be realized, thereby improving the obstacle avoidance effect of the medical logistics robot.

[0085] By performing three-dimensional mapping and motion modeling on the real-time obstacle image sequence based on the real-time point cloud sequence, the actual motion trajectory of the obstacle in the real-time path can be obtained, thereby improving the accuracy of the medical logistics robot's obstacle avoidance and the flexibility of path planning. By performing heuristic sampling and iterative path search, adaptive obstacle avoidance and path planning can be performed in combination with the predicted motion trajectory of the obstacle, and real-time path updates can be achieved through global evaluation optimization, thereby improving the flexibility of route planning. Therefore, the medical logistics robot route planning method and system based on adaptive feedback proposed in the present invention can solve the problem of low flexibility in route planning. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] Figure 1 A schematic diagram of a process flow of a medical logistics robot route planning method based on adaptive feedback provided by one embodiment of the present invention;

[0087] Figure 2 A schematic flow chart of a method for extracting a real-time obstacle image sequence provided by one embodiment of the present invention;

[0088] Figure 3 A schematic flow chart of a method for extracting a primary obstacle detection frame sequence provided by one embodiment of the present invention;

[0089] Figure 4 This is a functional module diagram of a medical logistics robot route planning system based on adaptive feedback provided by one embodiment of the present invention;

[0090] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0091] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0092] The embodiment of the present application provides a medical logistics robot route planning method based on adaptive feedback. The execution subject of the medical logistics robot route planning method based on adaptive feedback includes but is not limited to at least one of the electronic devices such as the server, the terminal, etc. that can be configured to execute the method provided by the embodiment of the present application. In other words, the medical logistics robot route planning method based on adaptive feedback can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc. The server can be an independent server, or it can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0093] Reference Figure 1 FIG2 is a flow chart of a medical logistics robot route planning method based on adaptive feedback according to an embodiment of the present invention. In this embodiment, the medical logistics robot route planning method based on adaptive feedback includes:

[0094] S1. Obtain a hospital static route map and real-time dynamic logistics information, and perform topology reconstruction and path planning on the hospital static route map based on the real-time dynamic logistics information to obtain a primary logistics route sequence.

[0095] In detail, the hospital static roadmap is a graphic representation of the planar or spatial layout of the hospital's internal structure and its fixed facilities. It is usually used to describe the spatial relationship between various functional areas, passages, rooms and facilities in the hospital, and mainly reflects the static infrastructure layout in the hospital that does not change over time.

[0096] Specifically, the real-time dynamic logistics information refers to real-time logistics-related dynamic information, such as the departure location, delivery location, opening status of the hospital's automatic doors, and the usage status of each elevator in the hospital. The hospital's static route map and real-time dynamic logistics information primary logistics route sequence can be obtained from the hospital's monitoring system.

[0097] In detail, the primary logistics route sequence is a route from the departure point to the delivery point composed of multiple consecutive primary logistics routes, and two consecutive primary logistics routes are connected end to end, that is, the end point of the previous primary logistics route is the starting point of the next primary logistics route.

[0098] In an embodiment of the present invention, the topology reconstruction and path planning of the hospital static route map based on the real-time dynamic logistics information to obtain a primary logistics route sequence includes:

[0099] Extracting a static node set and a static route set from the hospital static route map respectively;

[0100] Extracting a route attribute set corresponding to the static route set from the real-time dynamic logistics information;

[0101] Performing route feature extraction and route weight mapping on the route attribute set to obtain a route weight set;

[0102] Performing topology reconstruction based on the static node set and the static route set to obtain a primary route topology graph;

[0103] Performing path weighting on the primary route topology map using the route weight set to obtain a standard route topology map;

[0104] Extracting a logistics starting point and a logistics end point from the real-time dynamic logistics information;

[0105] Path planning is performed on the standard route topology map according to the logistics starting point and the logistics end point to obtain a primary logistics route sequence.

[0106] In detail, each static node in the static node set corresponds to an intersection of routes in the hospital static route map, and each static route in the static route set corresponds to a route of each section in the hospital static route map, that is, a route used to connect each static node.

[0107] Specifically, each route attribute in the route attribute set refers to attribute information of the corresponding static route in the static route set, such as the inclination angle of the route, the ground material, the waiting time of the elevator in the route, and other information.

[0108] In detail, the route feature extraction refers to extracting the attribute features corresponding to the route attributes. The route feature extraction can be performed using regular expressions or character matching methods. The route weight mapping refers to pre-setting weights for the corresponding route feature mapping, and taking the weighted average of each weight as the route weight. For example, the weight corresponding to the feature with an inclination angle of 0 is 1, and the weight corresponding to a smooth ground material is 0.8.

[0109] Specifically, the topology reconstruction based on the static node set and the static route set to obtain a primary route topology map includes: generating an adjacency matrix based on the static node set and the static route set; generating an adjacency route topology map based on the adjacency matrix; and performing redundancy optimization on the adjacency route topology map to obtain a primary route topology map.

[0110] In detail, the adjacency matrix is ​​a matrix used to represent the connection relationship between the static nodes in the static node set. When there is a connection relationship between the static nodes, that is, there is a corresponding static route, the corresponding element in the adjacency matrix is ​​1, otherwise it is 0.

[0111] Specifically, the adjacency route topology graph can be generated according to the adjacency matrix using graph theory library components such as NetworkX and igraph. The redundancy optimization refers to deleting isolated points or redundant edges in the adjacency route topology graph, thereby simplifying the graph structure. The node degree statistics method can be used to screen out isolated points, and the depth-first search method can be used to screen out redundant edges. The path weighting refers to multiplying the weight of each route in the route weight set by the length of the corresponding route in the primary route topology graph, and using the weighted route length to update the primary route topology graph, thereby obtaining the final standard route topology graph.

[0112] In detail, the logistics starting point refers to the departure point of this round of medical logistics robot logistics distribution in the real-time dynamic logistics information, and the logistics end point refers to the delivery point of this round of medical logistics robot logistics distribution in the real-time dynamic logistics information.

[0113] Specifically, the path planning is performed on the standard route topology map according to the logistics starting point and the logistics end point to obtain a primary logistics route sequence, including: extracting the starting node corresponding to the logistics starting point and the end node corresponding to the logistics end point from the standard route topology map respectively; extracting a standard node set and a standard route set from the standard route topology map; setting the cost value of the standard nodes in the standard node set except the starting node to infinity, and setting the cost value of the starting node to zero; using the Dijkstra algorithm to perform path planning on the standard route topology map according to the cost value to obtain a primary logistics route sequence from the starting node to the end node.

[0114] In an embodiment of the present invention, by topologically reconstructing and planning the path of the hospital static road map based on the real-time dynamic logistics information, a primary logistics route sequence is obtained, which can be combined with the complex road environment in the hospital to achieve global weighted path planning, thereby determining the primary robot driving path and improving the flexibility of path planning.

[0115] S2. Select primary logistics routes in the primary logistics route sequence one by one, perform path navigation and real-scene sensing monitoring according to the primary logistics routes, and obtain a real-scene point cloud sequence and a real-scene image sequence.

[0116] In detail, the path navigation refers to driving the medical logistics robot along the primary logistics route, and the real-scene sensing monitoring refers to using multiple sensors to obtain real-scene information of the surrounding area in real time, and the sensors include lidar and real-scene cameras.

[0117] In an embodiment of the present invention, the real-time point cloud sequence refers to a sequence of real-time point cloud data continuously acquired during the path navigation process, each real-time point cloud in the real-time point cloud sequence refers to point cloud data acquired in a corresponding time sequence frame, and the real-scene image sequence refers to a sequence composed of real-time environmental images continuously captured during the path navigation process, and each real-scene image in the real-scene image sequence refers to an environmental image acquired in a corresponding time sequence frame.

[0118] In an embodiment of the present invention, by performing path navigation and real-scene sensing monitoring according to the primary logistics route, a real-scene point cloud sequence and a real-scene image sequence are obtained, and point cloud data and real-scene images of the current driving path can be obtained, thereby realizing multi-dimensional detailed perception of the surrounding environment, obtaining more accurate environmental information, and improving the accuracy of adaptive feedback and path planning.

[0119] S3. Perform obstacle recognition and dynamic object tracking on the real scene image sequence to obtain a real-time obstacle image sequence.

[0120] In detail, each real-time obstacle image in the real-time obstacle image sequence corresponds to an image of an obstacle and a moving object in each real-scene image in the real-scene image sequence.

[0121] In the embodiment of the present invention, referring to Figure 2 As shown, the obstacle recognition and dynamic object tracking are performed on the real scene image sequence to obtain a real-time obstacle image sequence, including:

[0122] S21, performing edge detection and road fitting on the real scene image sequence to obtain real scene road edges;

[0123] S22, performing primary obstacle detection on the real scene image sequence to obtain a primary obstacle detection frame sequence;

[0124] S23. Perform motion analysis on the primary obstacle detection frame sequence to obtain an analyzed obstacle detection frame sequence;

[0125] S24, performing data association update on the primary obstacle detection frame sequence and the analysis obstacle detection frame sequence to obtain an associated obstacle detection frame sequence;

[0126] S25, performing obstacle screening on the associated obstacle detection frame sequence according to the real road edge to obtain a real-time obstacle detection frame sequence;

[0127] S26: Extracting a real-time obstacle image sequence from the real-scene image sequence using the real-time obstacle detection frame sequence.

[0128] In detail, the real-scene road edge refers to the edge of the road area in the current real-scene road corresponding to the real-scene image sequence. Edge detection can be performed using the Canny operator, Sobel operator or gradient histogram algorithm, and road fitting can be performed using algorithms such as polynomial fitting or least squares method.

[0129] Specifically, refer to Figure 3 As shown, the obstacle detection is performed on the real scene image sequence to obtain a primary obstacle detection frame sequence, including:

[0130] S31, selecting real scene images in the real scene image sequence one by one as target real scene images, performing adaptive residual convolution on the target real scene images, and obtaining residual real scene features;

[0131] S32, performing layer-by-layer multi-scale downsampling on the residual real scene features to obtain a multi-scale real scene feature set;

[0132] S33, performing feature slicing and local attention encoding on the multi-scale real scene feature set to obtain a local real scene feature set;

[0133] S34, performing global attention encoding on the local real scene feature set to obtain a global real scene feature set;

[0134] S35, performing multi-level feedforward activation and residual connection on the global real scene feature set to obtain an enhanced real scene feature set;

[0135] S36, performing layer-by-layer upsampling and obstacle feature activation on the enhanced real scene feature set to obtain a primary obstacle detection frame;

[0136] S37: Gather the activated obstacle detection frames corresponding to all target real scene images in the real scene image sequence into a primary obstacle detection frame sequence.

[0137] In detail, the adaptive residual convolution includes operations such as convolution, batch normalization, feature activation, adaptive convolution and residual connection. Feature activation can be performed using ReLU activation function or sigmoid activation function, and adaptive convolution can be performed using Involution operator. The residual real-scene feature combines the real-scene features before and after the residual connection and has more image feature details.

[0138] Specifically, the layer-by-layer multi-scale downsampling refers to reducing the scale of the residual real scene features layer by layer, downsampling the residual real scene features, and aggregating the sampling results of each layer and the residual real scene features into a multi-scale real scene feature set.

[0139] In detail, the feature slicing refers to dividing each multi-scale real scene feature in the multi-scale real scene feature set into real scene feature groups, which can be encoded using a local attention mechanism and encoded using a global attention mechanism.

[0140] Specifically, the step of performing layer-by-layer upsampling and obstacle feature activation on the enhanced real scene feature set to obtain a primary obstacle detection frame includes:

[0141] Sorting the enhanced real scene feature set in ascending order of feature size to obtain an enhanced real scene feature sequence;

[0142] Selecting the first enhanced real scene feature in the enhanced real scene feature sequence as the target first real scene feature, and selecting the enhanced real scene feature following the first real scene feature in the enhanced real scene feature sequence as the second real scene feature;

[0143] Using the characteristic size of the second real scene feature as the target characteristic size;

[0144] Upsampling the first real scene feature according to the target feature size to obtain a target upsampled real scene feature;

[0145] Performing feature fusion on the target upsampled real scene feature and the second real scene feature to obtain a target fusion feature;

[0146] Determining whether the second real scene feature is the last enhanced real scene feature in the enhanced real scene feature sequence;

[0147] If not, then using the target fusion feature to update the target first real scene feature, taking the enhanced real scene feature located after the second real scene feature in the enhanced real scene feature sequence as the second real scene feature, and returning to the step of taking the feature size of the second real scene feature as the target feature size;

[0148] If so, feature decoding and feature activation are performed on the target fusion feature to obtain a primary obstacle detection frame.

[0149] In detail, the feature size refers to the feature map size of each enhanced real scene feature in the enhanced real scene feature set, the feature fusion method can be a mean fusion method or a weighted fusion method, feature decoding can be performed using a convolution block, and feature activation can be performed using activation functions such as sigmoid or softmax.

[0150] Specifically, the performing data association updating on the primary obstacle detection frame sequence and the analysis obstacle detection frame sequence to obtain a real-time obstacle detection frame sequence includes:

[0151] selecting primary obstacle detection frames in the primary obstacle detection frame sequence one by one as target primary detection frames, and using analysis obstacle detection frames corresponding to the target primary detection frames in the analysis obstacle detection frame sequence as target analysis detection frames;

[0152] Extracting a primary obstacle image from the real scene image sequence according to the target primary detection frame, and extracting an analysis obstacle image from the real scene image sequence according to the target analysis detection frame;

[0153] extracting primary obstacle features from the primary obstacle image and extracting analytical obstacle features from the analytical obstacle image;

[0154] Calculating a detection frame distance between the target primary detection frame and the target analysis detection frame;

[0155] Calculating a characteristic distance between the primary obstacle feature and the analysis obstacle feature;

[0156] The position of the target analysis detection frame is updated according to the detection frame distance and the feature distance to obtain a real-time obstacle detection frame;

[0157] The real-time obstacle detection frames corresponding to all target analysis detection frames in the analysis obstacle detection frame sequence are aggregated into a real-time obstacle detection frame sequence.

[0158] In detail, the data association update algorithm can retain the correlation between features while determining the distance between features, thereby improving the accuracy of obstacle detection.

[0159] Specifically, updating the position of the target analysis detection frame according to the detection frame distance and the feature distance to obtain a real-time obstacle detection frame refers to using the Hungarian algorithm to perform target matching on the target primary detection frame and the target analysis detection frame according to the detection frame distance and the feature distance to obtain a matching result. For the corresponding target analysis detection frame whose matching result is a matching failure, the detection frame is updated to obtain a real-time obstacle detection frame.

[0160] In detail, the obstacle screening of the associated obstacle detection frame sequence according to the real-scene road edge to obtain a real-time obstacle detection frame sequence refers to aggregating the associated obstacle detection frames in the associated obstacle detection frame sequence within the area within the real-scene road edge as real-time obstacle detection frames into a real-time obstacle detection frame sequence.

[0161] In an embodiment of the present invention, by performing obstacle recognition and dynamic object tracking on the real-scene image sequence, obstacles that may hinder the medical logistics robot in the real-scene area can be effectively identified, and the motion tracking of the obstacles can be achieved, thereby improving the obstacle avoidance effect of the medical logistics robot.

[0162] S4. Perform three-dimensional mapping and motion modeling on the real-time obstacle image sequence according to the real-time point cloud sequence to obtain an obstacle motion path.

[0163] In detail, the obstacle motion path refers to the three-dimensional moving path of the obstacle in the real-time point cloud sequence.

[0164] In an embodiment of the present invention, performing three-dimensional mapping and motion modeling on the real-time obstacle image sequence according to the real-time point cloud sequence to obtain the obstacle motion path includes:

[0165] Extracting a point cloud feature point sequence from the real-time point cloud sequence, and extracting an obstacle feature point sequence from the real-time obstacle image sequence;

[0166] Performing feature point matching and coordinate alignment on the obstacle feature point sequence according to the point cloud feature point sequence to obtain a coordinate mapping relationship;

[0167] Mapping the real-time obstacle image sequence to the real-time point cloud sequence according to the coordinate mapping relationship to obtain a mapped point cloud sequence;

[0168] performing point cloud denoising on the mapped point cloud sequence to obtain an obstacle point cloud sequence;

[0169] Performing three-dimensional reconstruction on the obstacle point cloud sequence to obtain an obstacle model sequence;

[0170] performing motion estimation on the obstacle model sequence to obtain an obstacle motion model;

[0171] Path mapping and path smoothing are performed on the obstacle motion model to obtain an obstacle motion path.

[0172] In detail, a corner detection algorithm or a feature point detection algorithm such as SIFT can be used to extract a point cloud feature point sequence and an obstacle feature point sequence. The feature point matching refers to matching the feature matching values ​​between the features corresponding to each point cloud feature point in the point cloud feature point sequence and the features corresponding to each obstacle feature point in the obstacle feature point sequence one by one. The cosine feature distance algorithm can be used for feature point matching. The coordinate alignment refers to calculating the coordinate mapping relationship between the successfully matched point cloud feature points and the corresponding obstacle feature points.

[0173] In detail, the point cloud filtering algorithm can be combined with Gaussian filtering to achieve noise removal filtering of the point cloud, while making the filtering algorithm more flexible and able to adapt to more diverse point cloud data characteristics.

[0174] Specifically, three-dimensional reconstruction can be performed using voxelization or surface reconstruction algorithms. Each obstacle model in the obstacle model sequence corresponds to a three-dimensional model of an obstacle in a real road within a time frame. Motion estimation can be performed using a nonlinear motion model, a rigid body motion model, or a linear motion model to obtain an obstacle motion model.

[0175] In detail, the path mapping refers to using a set of time-related coordinate sequences to represent the corresponding obstacle motion trajectory, and path smoothing can be performed using methods such as B-Spline smoothing or Kalman filtering.

[0176] In an embodiment of the present invention, by performing three-dimensional mapping and motion modeling on the real-time obstacle image sequence based on the real-time point cloud sequence, the actual motion trajectory of the obstacle in the real-time path can be obtained, thereby improving the accuracy of the medical logistics robot's obstacle avoidance and improving the flexibility of path planning.

[0177] S5. Perform heuristic sampling and iterative path search on the primary logistics route according to the obstacle movement path to obtain a standard logistics route, and perform global evaluation and optimization on the primary logistics route sequence according to the real-time dynamic logistics information and the standard logistics route to obtain a standard logistics route sequence.

[0178] In detail, the standard logistics route refers to the shortest route within the section corresponding to the primary logistics route that can avoid obstacles and reach the end point of the section.

[0179] In an embodiment of the present invention, the heuristic sampling and iterative path search of the primary logistics route according to the obstacle movement path to obtain the standard logistics route includes:

[0180] Obtaining a route map of the primary logistics route;

[0181] Dividing the route map into grids according to a preset robot size to obtain a logistics map grid;

[0182] Extracting a route end point and a route starting point from the primary logistics route;

[0183] Marking the points of the logistics map grid according to the route end point and the route starting point to obtain a marked map grid;

[0184] De-weighting the marked map grid according to the route endpoint to obtain a primary sampling grid;

[0185] Performing weighted sampling on the primary sampling grid according to the obstacle motion path to obtain a standard sampling grid;

[0186] An iterative path search is performed on the standard sampling grid according to the obstacle movement path to obtain a standard logistics route.

[0187] Specifically, the route map refers to the road map corresponding to the primary logistics route, and the route map includes information such as the length and width of the route. The robot size refers to the footprint of the medical logistics robot. The route map is gridded according to the preset robot size to obtain the logistics map grid, which means generating the corresponding external minimum grid according to the robot size, and evenly gridding the route map according to the external minimum grid.

[0188] In detail, the route end point refers to the end node corresponding to the primary logistics route, the route starting point refers to the starting node corresponding to the primary logistics route, and the point marking refers to marking the route end point and the route starting point on the logistics map grid.

[0189] Specifically, the down-sampling refers to reducing the weight value of the neighborhood grid of the starting point of the route in the marked map grid, and the weighted sampling refers to increasing the grid weight value corresponding to the obstacle motion path in the marked map grid, wherein the initial grid weights of each grid in the marked map grid are the same.

[0190] Specifically, performing an iterative path search on the standard sampling grid according to the obstacle movement path to obtain a standard logistics route includes:

[0191] Initializing a particle group, selecting particles in the particle group one by one as target particles, and using the starting point of the route in the standard sampling grid as the initial position of the target particle;

[0192] Calculating a grid movement interval based on a preset robot movement speed and the standard sampling grid;

[0193] Sampling the obstacle motion path frame by frame according to the grid movement interval to obtain a target frame obstacle grid;

[0194] Performing safety area screening on the standard sampling grid according to the target frame obstacle grid to obtain a safe sampling grid;

[0195] updating the position of the target particle in the safety sampling grid one by one to obtain a target update position, and generating a moving route of the target particle according to all target update positions and the initial position;

[0196] Determining whether the target update position is located at a route end point in the standard sampling grid graph;

[0197] If not, returning to the step of sampling the obstacle motion path frame by frame according to the grid movement interval to obtain the target frame obstacle grid;

[0198] If yes, the moving route is used as the initial search route of the target particle;

[0199] Aggregating the initial search routes of all target particles in the particle group into an initial search route set;

[0200] The initial search route set is screened for the shortest route to obtain a standard logistics route.

[0201] Specifically, the robot movement speed refers to the movement speed of the medical logistics robot, and the medical logistics robot moves at an approximately uniform speed. The grid movement interval is obtained by dividing the average of the side length and diagonal of the standard sampling grid by the robot movement speed.

[0202] In detail, the safe area screening refers to screening out the grids in the standard sampling grid except the target frame obstacle grid as the safe sampling grid, and updating the position of the target particle in the safe sampling grid one by one to obtain the target updated position means moving the target particle to an adjacent grid or keeping it still, wherein, in the case of being still, the target particle can only wait continuously in the same grid for the time interval of three grid movements.

[0203] Specifically, generating the moving route of the target particle according to all target update positions and the initial position means taking the time-ordered path of all target update positions starting from the initial position as the moving route.

[0204] In detail, the shortest route screening refers to screening out the route with the smallest weighted value among all the initial search routes in the primary search route set as the standard logistics route, wherein the weighted value refers to the sum of the grid weight values ​​of each grid.

[0205] Specifically, the global evaluation and optimization of the primary logistics route sequence is performed according to the real-time dynamic logistics information and the standard logistics route to obtain a standard logistics route sequence, including:

[0206] Determining whether the primary logistics route corresponding to the standard logistics route is the last primary logistics route in the primary logistics route sequence;

[0207] If not, then perform a time analysis on the standard logistics route to obtain the standard logistics time;

[0208] The real-time dynamic logistics information is updated according to the standard logistics duration to obtain standard dynamic logistics information;

[0209] Perform route overlay on the hospital static route map according to the standard logistics route to obtain a covered static route map;

[0210] Performing topology reconstruction and path planning on the covered static route map according to the standard dynamic logistics information to obtain an updated logistics route sequence;

[0211] Using the updated logistics route sequence to update the primary logistics route sequence, and returning to the step of selecting primary logistics routes in the primary logistics route sequence one by one;

[0212] If so, the standard logistics routes corresponding to all primary logistics routes in the primary logistics route sequence are aggregated into a standard logistics route sequence.

[0213] In detail, the standard dynamic logistics information refers to the actual dynamic logistics information obtained after the time period corresponding to the standard logistics duration, and the route coverage refers to screening out and covering the routes corresponding to the standard logistics routes in the hospital static route map.

[0214] Specifically, the topology reconstruction and path planning methods are consistent with the methods in the above step S1 and will not be repeated here. By performing global evaluation optimization, real-time path planning updates can be achieved, thereby meeting the real-time requirements of hospital logistics and improving the flexibility of path planning.

[0215] In the embodiment of the present invention, by performing heuristic sampling and iterative path search, adaptive obstacle avoidance and path planning can be performed in combination with the predicted motion trajectory of the obstacle, and real-time path updates can be achieved through global evaluation optimization, thereby improving the flexibility of route planning.

[0216] The present invention obtains a primary logistics route sequence by topologically reconstructing and path planning the hospital static route map according to the real-time dynamic logistics information. It can realize global weighted path planning in combination with the complex road environment in the hospital, thereby determining the primary robot driving path and improving the flexibility of path planning. By performing path navigation and real-scene sensing monitoring according to the primary logistics route, a real-scene point cloud sequence and a real-scene image sequence are obtained, and point cloud data and real-scene images of the current driving path can be obtained, thereby realizing multi-dimensional detailed perception of the surrounding environment, obtaining more accurate environmental information, and improving the accuracy of adaptive feedback and path planning. By performing obstacle recognition and dynamic object tracking on the real-scene image sequence, obstacles that may hinder the medical logistics robot in the real-scene area can be effectively identified, and the motion tracking of the obstacles can be realized, thereby improving the obstacle avoidance effect of the medical logistics robot.

[0217] By performing three-dimensional mapping and motion modeling on the real-time obstacle image sequence based on the real-time point cloud sequence, the actual motion trajectory of the obstacle in the real-time path can be obtained, thereby improving the accuracy of the medical logistics robot's obstacle avoidance and the flexibility of path planning. By performing heuristic sampling and iterative path search, adaptive obstacle avoidance and path planning can be performed in combination with the predicted motion trajectory of the obstacle, and real-time path updates can be achieved through global evaluation optimization, thereby improving the flexibility of route planning. Therefore, the medical logistics robot route planning method based on adaptive feedback proposed in the present invention can solve the problem of low flexibility in route planning.

[0218] like Figure 4 , which is a functional module diagram of a medical logistics robot route planning system based on adaptive feedback provided by one embodiment of the present invention.

[0219] The adaptive feedback-based medical logistics robot route planning system 100 described in the present invention can be installed in an electronic device. Depending on the functions implemented, the adaptive feedback-based medical logistics robot route planning system 100 can include a topology planning module 101, a real-time monitoring module 102, an obstacle tracking module 103, a motion modeling module 104, and a global optimization module 105. The modules described in the present invention, also known as units, refer to a series of computer program segments that can be executed by an electronic device processor and can perform a fixed function, and are stored in the memory of the electronic device.

[0220] In this embodiment, the functions of each module / unit are as follows:

[0221] The topology planning module 101 is used to obtain a hospital static route map and real-time dynamic logistics information, and perform topology reconstruction and path planning on the hospital static route map based on the real-time dynamic logistics information to obtain a primary logistics route sequence;

[0222] The real-scene monitoring module 102 is configured to select primary logistics routes from the primary logistics route sequence one by one, perform path navigation and real-scene sensing monitoring according to the primary logistics routes, and obtain a real-scene point cloud sequence and a real-scene image sequence;

[0223] The obstacle tracking module 103 is used to perform obstacle recognition and dynamic object tracking on the real scene image sequence to obtain a real-time obstacle image sequence;

[0224] The motion modeling module 104 is configured to perform three-dimensional mapping and motion modeling on the real-time obstacle image sequence according to the real-time point cloud sequence to obtain an obstacle motion path, wherein the performing three-dimensional mapping and motion modeling on the real-time obstacle image sequence according to the real-time point cloud sequence to obtain the obstacle motion path comprises: extracting a point cloud feature point sequence from the real-time point cloud sequence and extracting an obstacle feature point sequence from the real-time obstacle image sequence; performing feature point matching and coordinate alignment on the obstacle feature point sequence according to the point cloud feature point sequence to obtain a coordinate mapping relationship; mapping the real-time obstacle image sequence to the real-time point cloud sequence according to the coordinate mapping relationship to obtain a mapped point cloud sequence; performing point cloud denoising on the mapped point cloud sequence to obtain an obstacle point cloud sequence; performing three-dimensional reconstruction on the obstacle point cloud sequence to obtain an obstacle model sequence; performing motion estimation on the obstacle model sequence to obtain an obstacle motion model; and performing path mapping and path smoothing on the obstacle motion model to obtain an obstacle motion path.

[0225] The global optimization module 105 is used to perform heuristic sampling and iterative path search on the primary logistics route according to the obstacle movement path to obtain a standard logistics route, and to perform global evaluation and optimization on the primary logistics route sequence according to the real-time dynamic logistics information and the standard logistics route to obtain a standard logistics route sequence.

[0226] In detail, each module described in the medical logistics robot route planning system 100 based on adaptive feedback in the embodiment of the present invention adopts the same Figure 1 The same technical means are used as the medical logistics robot route planning method based on adaptive feedback described in , and can produce the same technical effects, so I will not go into details here.

[0227] In the several embodiments provided herein, it should be understood that the disclosed devices, systems, and methods may be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the module division is merely a logical functional division, and actual implementation may employ other division methods.

[0228] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed across multiple network elements. Some or all of the modules may be selected to achieve the purpose of the solution of this embodiment according to actual needs.

[0229] In addition, the functional modules in various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or hardware plus software functional modules.

[0230] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0231] Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the claims are intended to be embraced therein. Any reference to a figure in a claim should not be construed as limiting the claim to which it relates.

[0232] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, techniques, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results.

[0233] Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or systems described in a system embodiment may also be implemented by a single unit or system through software or hardware. Terms such as first and second are used to indicate names and do not imply any particular order.

[0234] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A medical logistics robot route planning method based on adaptive feedback, characterized in that: The method comprises: Obtaining a hospital static route map and real-time dynamic logistics information, and performing topological reconstruction and path planning on the hospital static route map based on the real-time dynamic logistics information to obtain a primary logistics route sequence; Selecting primary logistics routes from the primary logistics route sequence one by one, performing path navigation and real-scene sensing monitoring according to the primary logistics routes, and obtaining a real-scene cloud sequence and a real-scene image sequence; Performing obstacle recognition and dynamic object tracking on the real scene image sequence to obtain a real scene obstacle image sequence; Performing three-dimensional mapping and motion modeling on the real-scene obstacle image sequence according to the real-scene point cloud sequence to obtain an obstacle motion path, wherein performing three-dimensional mapping and motion modeling on the real-scene obstacle image sequence according to the real-scene point cloud sequence to obtain the obstacle motion path includes: extracting a point cloud feature point sequence from the real-scene point cloud sequence, and extracting an obstacle feature point sequence from the real-scene obstacle image sequence; performing feature point matching and coordinate alignment on the obstacle feature point sequence according to the point cloud feature point sequence to obtain a coordinate mapping relationship; mapping the real-scene obstacle image sequence to the real-scene point cloud sequence according to the coordinate mapping relationship to obtain a mapped point cloud sequence; performing point cloud denoising on the mapped point cloud sequence to obtain an obstacle point cloud sequence; performing three-dimensional reconstruction on the obstacle point cloud sequence to obtain an obstacle model sequence; performing motion estimation on the obstacle model sequence to obtain an obstacle motion model; performing path mapping and path smoothing on the obstacle motion model to obtain an obstacle motion path; The primary logistics route is subjected to heuristic sampling and iterative path search according to the obstacle motion path to obtain a standard logistics route, and the primary logistics route sequence is globally evaluated and optimized according to the real-time dynamic logistics information and the standard logistics route to obtain a standard logistics route sequence; wherein, the heuristic sampling and iterative path search of the primary logistics route according to the obstacle motion path to obtain the standard logistics route includes: obtaining a route map of the primary logistics route; gridding the route map according to a preset robot size to obtain a logistics map grid; extracting a route end point and a route starting point from the primary logistics route; marking the logistics map grid according to the route end point and the route starting point to obtain a marked map grid; performing down-weighted sampling on the marked map grid according to the route end point to obtain a primary sampling grid; performing weighted sampling on the primary sampling grid according to the obstacle motion path to obtain a standard sampling grid; performing iterative path search on the standard sampling grid according to the obstacle motion path to obtain a standard logistics route.

2. The medical logistics robot route planning method based on adaptive feedback according to claim 1, characterized in that: The topology reconstruction and path planning of the hospital static route map are performed according to the real-time dynamic logistics information to obtain a primary logistics route sequence, including: Extracting a static node set and a static route set from the hospital static route map respectively; Extracting a route attribute set corresponding to the static route set from the real-time dynamic logistics information; Performing route feature extraction and route weight mapping on the route attribute set to obtain a route weight set; Performing topology reconstruction based on the static node set and the static route set to obtain a primary route topology graph; Performing path weighting on the primary route topology map using the route weight set to obtain a standard route topology map; Extracting a logistics starting point and a logistics end point from the real-time dynamic logistics information; Path planning is performed on the standard route topology map according to the logistics starting point and the logistics end point to obtain a primary logistics route sequence.

3. The medical logistics robot route planning method based on adaptive feedback according to claim 1, characterized in that: The performing obstacle recognition and dynamic object tracking on the real scene image sequence to obtain a real scene obstacle image sequence includes: Performing edge detection and road fitting on the real scene image sequence to obtain real scene road edges; Performing primary obstacle detection on the real scene image sequence to obtain a primary obstacle detection frame sequence; Performing motion analysis on the primary obstacle detection frame sequence to obtain an analyzed obstacle detection frame sequence; performing data association and updating on the primary obstacle detection frame sequence and the analysis obstacle detection frame sequence to obtain an associated obstacle detection frame sequence; Performing obstacle screening on the associated obstacle detection frame sequence according to the real-scene road edge to obtain a real-scene obstacle detection frame sequence; A real scene obstacle image sequence is extracted from the real scene image sequence using the real scene obstacle detection frame sequence.

4. The medical logistics robot route planning method based on adaptive feedback according to claim 3, characterized in that: The performing obstacle detection on the real scene image sequence to obtain a primary obstacle detection frame sequence includes: Selecting real scene images in the real scene image sequence one by one as target real scene images, performing adaptive residual convolution on the target real scene images, and obtaining residual real scene features; Performing layer-by-layer multi-scale downsampling on the residual real scene features to obtain a multi-scale real scene feature set; Performing feature slicing and local attention encoding on the multi-scale real scene feature set to obtain a local real scene feature set; Performing global attention encoding on the local real scene feature set to obtain a global real scene feature set; Performing multi-level feedforward activation and residual connection on the global real scene feature set to obtain an enhanced real scene feature set; Performing layer-by-layer upsampling and obstacle feature activation on the enhanced real scene feature set to obtain a primary obstacle detection frame; The activated obstacle detection frames corresponding to all target real scene images in the real scene image sequence are collected into a primary obstacle detection frame sequence.

5. The medical logistics robot route planning method based on adaptive feedback according to claim 4, characterized in that: The step of upsampling the enhanced real scene feature set layer by layer and activating obstacle features to obtain a primary obstacle detection frame includes: Sorting the enhanced real scene feature set in ascending order of feature size to obtain an enhanced real scene feature sequence; Selecting the first enhanced real scene feature in the enhanced real scene feature sequence as the target first real scene feature, and selecting the enhanced real scene feature following the first real scene feature in the enhanced real scene feature sequence as the second real scene feature; Using the characteristic size of the second real scene feature as the target characteristic size; Upsampling the first real scene feature according to the target feature size to obtain a target upsampled real scene feature; Performing feature fusion on the target upsampled real scene feature and the second real scene feature to obtain a target fusion feature; Determining whether the second real scene feature is the last enhanced real scene feature in the enhanced real scene feature sequence; If not, then using the target fusion feature to update the target first real scene feature, taking the enhanced real scene feature located after the second real scene feature in the enhanced real scene feature sequence as the second real scene feature, and returning to the step of taking the feature size of the second real scene feature as the target feature size; If so, feature decoding and feature activation are performed on the target fusion feature to obtain a primary obstacle detection frame.

6. The medical logistics robot route planning method based on adaptive feedback according to claim 3, characterized in that: The step of performing data association updating on the primary obstacle detection frame sequence and the analysis obstacle detection frame sequence to obtain a real scene obstacle detection frame sequence includes: selecting primary obstacle detection frames in the primary obstacle detection frame sequence one by one as target primary detection frames, and using analysis obstacle detection frames corresponding to the target primary detection frames in the analysis obstacle detection frame sequence as target analysis detection frames; Extracting a primary obstacle image from the real scene image sequence according to the target primary detection frame, and extracting an analysis obstacle image from the real scene image sequence according to the target analysis detection frame; extracting primary obstacle features from the primary obstacle image and extracting analytical obstacle features from the analytical obstacle image; Calculating a detection frame distance between the target primary detection frame and the target analysis detection frame; Calculate the characteristic distance between the primary obstacle feature and the analysis obstacle feature: The target analysis detection frame is updated according to the detection frame distance and the feature distance to obtain a real scene obstacle detection frame; The real scene obstacle detection frames corresponding to all target analysis detection frames in the analysis obstacle detection frame sequence are aggregated into a real scene obstacle detection frame sequence.

7. The medical logistics robot route planning method based on adaptive feedback according to claim 1, characterized in that: The iterative path search is performed on the standard sampling grid according to the obstacle movement path to obtain a standard logistics route, including: Initializing a particle group, selecting particles in the particle group one by one as target particles, and using the starting point of the route in the standard sampling grid as the initial position of the target particle; Calculating a grid movement interval based on a preset robot movement speed and the standard sampling grid; Sampling the obstacle motion path frame by frame according to the grid movement interval to obtain a target frame obstacle grid; Performing safety area screening on the standard sampling grid according to the target frame obstacle grid to obtain a safe sampling grid; updating the position of the target particle in the safety sampling grid one by one to obtain a target update position, and generating a moving route of the target particle according to all target update positions and the initial position; Determining whether the target update position is located at a route end point in the standard sampling grid graph; If not, returning to the step of sampling the obstacle motion path frame by frame according to the grid movement interval to obtain the target frame obstacle grid; If yes, the moving route is used as the initial search route of the target particle; Aggregating the initial search routes of all target particles in the particle group into an initial search route set; The initial search route set is screened for the shortest route to obtain a standard logistics route.

8. The medical logistics robot route planning method based on adaptive feedback according to claim 1, characterized in that: The globally evaluating and optimizing the primary logistics route sequence according to the real-time dynamic logistics information and the standard logistics route to obtain a standard logistics route sequence includes: Determining whether the primary logistics route corresponding to the standard logistics route is the last primary logistics route in the primary logistics route sequence; If not, then perform a time analysis on the standard logistics route to obtain the standard logistics time; The real-time dynamic logistics information is updated according to the standard logistics duration to obtain standard dynamic logistics information; Perform route overlay on the hospital static route map according to the standard logistics route to obtain a covered static route map; Performing topology reconstruction and path planning on the covered static route map according to the standard dynamic logistics information to obtain an updated logistics route sequence; Using the updated logistics route sequence to update the primary logistics route sequence, and returning to the step of selecting primary logistics routes in the primary logistics route sequence one by one; If so, the standard logistics routes corresponding to all primary logistics routes in the primary logistics route sequence are aggregated into a standard logistics route sequence.

9. A medical logistics robot route planning system based on adaptive feedback, characterized in that: The system comprises: A topology planning module is used to obtain a hospital static route map and real-time dynamic logistics information, and to perform topology reconstruction and path planning on the hospital static route map based on the real-time dynamic logistics information to obtain a primary logistics route sequence; A real-scene monitoring module is used to select primary logistics routes from the primary logistics route sequence one by one, perform path navigation and real-scene sensing monitoring according to the primary logistics routes, and obtain a real-scene point cloud sequence and a real-scene image sequence; An obstacle tracking module is used to perform obstacle recognition and dynamic object tracking on the real scene image sequence to obtain a real scene obstacle image sequence; a motion modeling module, configured to perform three-dimensional mapping and motion modeling on the real-scene obstacle image sequence according to the real-scene point cloud sequence to obtain an obstacle motion path, wherein the performing three-dimensional mapping and motion modeling on the real-scene obstacle image sequence according to the real-scene point cloud sequence to obtain the obstacle motion path comprises: extracting a point cloud feature point sequence from the real-scene point cloud sequence and extracting an obstacle feature point sequence from the real-scene obstacle image sequence; performing feature point matching and coordinate alignment on the obstacle feature point sequence according to the point cloud feature point sequence to obtain a coordinate mapping relationship; mapping the real-scene obstacle image sequence to the real-scene point cloud sequence according to the coordinate mapping relationship to obtain a mapped point cloud sequence; performing point cloud denoising on the mapped point cloud sequence to obtain an obstacle point cloud sequence; performing three-dimensional reconstruction on the obstacle point cloud sequence to obtain an obstacle model sequence; performing motion estimation on the obstacle model sequence to obtain an obstacle motion model; and performing path mapping and path smoothing on the obstacle motion model to obtain an obstacle motion path; A global optimization module is used to perform heuristic sampling and iterative path search on the primary logistics route according to the obstacle motion path to obtain a standard logistics route, and to perform global evaluation and optimization on the primary logistics route sequence according to the real-time dynamic logistics information and the standard logistics route to obtain a standard logistics route sequence; wherein, the heuristic sampling and iterative path search on the primary logistics route according to the obstacle motion path to obtain the standard logistics route includes: obtaining a route map of the primary logistics route; gridding the route map according to a preset robot size to obtain a logistics map grid; extracting a route end point and a route starting point from the primary logistics route; marking the logistics map grid according to the route end point and the route starting point to obtain a marked map grid; performing down-weighted sampling on the marked map grid according to the route end point to obtain a primary sampling grid; performing weighted sampling on the primary sampling grid according to the obstacle motion path to obtain a standard sampling grid; performing iterative path search on the standard sampling grid according to the obstacle motion path to obtain a standard logistics route.

Images