Obstacle avoidance method and system of medical displacement machine

Abstract

The invention discloses an obstacle avoidance method and system for a medical displacement machine, and the method comprises the steps: collecting and processing static and dynamic obstacle information in an environment, and obtaining first precise obstacle information; classifying the first accurate obstacle information through a convolutional neural network, and recording a first dynamic motion track; learning a dynamic obstacle motion mode by using a recurrent neural network, and outputting a predicted dynamic obstacle trajectory; a static path planning algorithm is adopted on a known map, and a preliminary path from a starting point to a target point is found in combination with the predicted dynamic obstacle trajectory; when it is detected that the moving direction or speed change of the dynamic obstacle is second accurate obstacle information, the second accurate obstacle information is used for outputting and updating the dynamic obstacle track; and outputting an obstacle avoidance path according to the updated dynamic obstacle trajectory. When the moving direction or speed change of the dynamic obstacle is detected, new obstacle information is immediately used to update the trajectory of the dynamic obstacle, and the path is re-planned, so that the real-time performance and effectiveness of obstacle avoidance are ensured.

Description

Technical Field

[0001] The present invention relates to the field of path navigation technology, and in particular to an obstacle avoidance method and system for a medical transfer machine. Background Art

[0002] In the medical field, with the development of technology, there are higher requirements for patient care and treatment efficiency. As an auxiliary device, medical transfer machines play an important role in helping patients with limited mobility to transfer positions. Ensuring that patients are safely and efficiently transferred from one location to another is a vital task, especially in large hospitals or nursing homes. The frequent demand for patient transfers has brought considerable pressure to medical staff.

[0003] Hospitals and nursing homes are complex indoor environments. Their unique spatial layout, crowd density, equipment distribution and other factors bring many technical challenges to the path planning of the lift. At present, the path planning of the lift is based on building a map to find the optimal path, but this method is only applicable to known environments. When pedestrians suddenly change the direction or speed of movement, especially when turning, going uphill or downhill, encountering obstacles will cause the lift to slip or change speed too much, affecting the stability of the lift and causing discomfort or danger to the patient. Summary of the invention

[0004] The present application provides an obstacle avoidance method and system for a medical transfer machine, which solves the technical problem in the prior art that the transfer machine has poor stability when encountering pedestrians who suddenly change the moving direction or speed.

[0005] The present application provides an obstacle avoidance method for a medical transfer machine, comprising the following steps:

[0006] Step S1: collecting static and dynamic obstacle information in the environment in real time through at least one laser sensor, at least one monocular camera and at least one depth camera;

[0007] Step S2: Using Kalman filtering to process static and dynamic obstacle information from the laser sensor, the monocular camera, and the depth camera to obtain first precise obstacle information;

[0008] Step S3: classifying the first precise obstacle information through a convolutional neural network to obtain precise dynamic obstacle information and precise static obstacle information, continuously tracking the dynamic obstacle, and recording the first dynamic motion trajectory;

[0009] Step S4: Obtain the first dynamic motion trajectory information from step S3, use a recurrent neural network to learn the dynamic obstacle motion pattern, and output a predicted dynamic obstacle trajectory;

[0010] Step S5: Using a static path planning algorithm on a known map, combined with the predicted dynamic obstacle trajectory, to find a preliminary path from the starting point to the target point;

[0011] Step S6: when it is detected that the moving direction or speed of the dynamic obstacle changes to the second precise obstacle information, the second precise obstacle information is immediately used to output and update the dynamic obstacle trajectory;

[0012] Step S7: Output the obstacle avoidance path according to the updated dynamic obstacle trajectory.

[0013] In some embodiments, in step S2, the static and dynamic obstacle information from the laser sensor, the monocular camera, and the depth camera are processed using Kalman filtering to obtain the first precise obstacle information, including:

[0014] Step S21: deploying laser sensors, monocular cameras and depth cameras at different positions on the shifter to cover the maximum range of the surrounding environment, collecting static and dynamic obstacle information in the environment in real time, determining system state variables, and initializing the covariance matrix;

[0015] Step S22: performing time synchronization processing on the information collected by the laser sensor, the monocular camera and the depth camera to remove noise data;

[0016] Step S23: state prediction, at each time step, the state transfer equation is used to predict the static and dynamic obstacle information at the current moment according to the static and dynamic obstacle information at the previous moment, and converted into predicted state variables through the observation equation;

[0017] Calculate the predicted covariance matrix based on the state transfer equation and the covariance matrix of the previous moment;

[0018] Step S24: Obtain the static and dynamic obstacle information at the current moment, convert it into observation state variables, and update the observation covariance matrix according to the observation state variables;

[0019] Step S25: Calculate the Kalman gain based on the prediction covariance matrix and the observation covariance matrix;

[0020] Step S26: obtaining updated predicted state variables according to the Kalman gain, predicted state variables and observed state variables, and obtaining an updated covariance matrix. The above steps are repeated at each time step.

[0021] In some embodiments, the static and dynamic obstacle information includes the location, size, speed and character status of the obstacle, and the character status includes the direction of the feet, the degree of leg bending, facial expression and line of sight direction.

[0022] In some embodiments, if the Kalman gain approaches 0, the reliability of the observation value at the current moment is low, and the predicted state variable is selected;

[0023] If the Kalman gain tends to 1, the reliability of the observation value at the current moment is high, and the observed state variable is selected.

[0024] In some embodiments, step S6: when it is detected that the moving direction or speed of the dynamic obstacle changes to the second precise obstacle information, immediately using the second precise obstacle information to output and update the dynamic obstacle trajectory includes:

[0025] Step S61: continuously tracking the second precise obstacle information through a convolutional neural network and recording a second dynamic motion trajectory;

[0026] Step S62: Use a recurrent neural network to learn the second dynamic motion trajectory and output an updated dynamic obstacle trajectory.

[0027] In some embodiments, the step S7: outputting an obstacle avoidance path according to the updated dynamic obstacle trajectory includes:

[0028] Step S71: Based on the updated dynamic obstacle trajectory information, path planning is performed using the Dijkstra algorithm;

[0029] Step S72: Setting a safety distance threshold,

[0030] When the distance between the lift and the dynamic obstacle is close to the threshold and can be avoided in time, the speed of the lift is adjusted according to the speed at which the dynamic obstacle approaches the lift;

[0031] When the distance between the lift and a dynamic obstacle approaches the threshold and the lift cannot avoid the obstacle in time, the lift will decelerate immediately until it stops completely and continue to move forward after the obstacle is cleared.

[0032] In some embodiments, step S73: in step S72, when the distance between the transfer machine and the dynamic obstacle is greater than the threshold, the transfer machine maintains the current speed or accelerates to enter the obstacle avoidance path;

[0033] When the distance between the transfer machine and the dynamic obstacle is less than or equal to the threshold, the transfer machine is controlled to decelerate, and the deceleration amplitude A of the distance between the transfer machine and the dynamic obstacle is calculated.

[0034] A=k×(Ll)

[0035] Where k is the adjustment coefficient, L is the safety distance threshold, and l is the distance between the lift and the dynamic obstacle;

[0036] Calculate the target speed V based on the deceleration amplitude A target ,

[0037] Vtarget =V current -A

[0038] Among them, V current is the current speed of the lift,

[0039] If V target is less than zero, then set V target =0, the lift stops moving;

[0040] Set the speed change rate ΔV. The speed adjustment frequency should be consistent with the update frequency of static and dynamic obstacle information.

[0041] V next -V current -ΔV

[0042] If V next Greater than V target , then set V next =V target ,

[0043] V next is the lower speed of the lift.

[0044] In some embodiments, step S6 is further included after step S5. * : Determine the coverage of the laser sensor, monocular camera and depth camera. When the lift approaches a corner, slow down the lift and control the lift to move forward in a zigzag or wavy manner.

[0045] Use a monocular camera and a depth camera to collect image information on both sides of the corner, and use a crowd detection algorithm to calculate the density of people on both sides of the corner.

[0046] According to the detected crowd density, the lift is controlled to choose the side with less crowd as the driving direction.

[0047] If the density of people on both sides of the corner is similar, the lift is controlled to move in the middle.

[0048] A path planning system for a medical transfer machine, applying the method described in any one of the above technical solutions, the system comprising:

[0049] A sensing module, including at least one laser sensor, at least one monocular camera and at least one depth camera, for collecting static and dynamic obstacle information in the environment in real time;

[0050] A data processing module is used to process the data information from the sensor module and output the predicted dynamic obstacle trajectory;

[0051] The path planning module is used to plan the path based on the output prediction of dynamic obstacle trajectories;

[0052] A speed control module adjusts the speed of the lift according to the distance between the dynamic obstacle and the lift;

[0053] The corner obstacle avoidance module calculates the density of human traffic through the crowd detection algorithm and switches the movement route of the lift;

[0054] The control system controls the movement of the lift according to the planned path planning information.

[0055] In some embodiments, the laser sensor and the monocular camera are installed at two ends of the same side of the transfer machine, and the depth camera is installed at the head of the bed of the transfer machine.

[0056] The beneficial effects of this application are as follows:

[0057] Through multi-sensor real-time acquisition of static and dynamic obstacle information in the environment, CNN classifies the first accurate obstacle information to obtain accurate dynamic obstacle information and accurate static obstacle information, which improves the accuracy of object detection. RNN is used to learn the movement mode of dynamic obstacles and output the predicted dynamic obstacle trajectory, which improves the accuracy of dynamic obstacle prediction and enables the system to make obstacle avoidance decisions in advance. When the moving direction or speed change of a dynamic obstacle is detected, the dynamic obstacle trajectory is immediately updated with the new obstacle information and the path is replanned to ensure the real-time and effectiveness of obstacle avoidance.

[0058] When the distance between the lift and a dynamic obstacle approaches the threshold, the system can dynamically adjust the speed of the lift according to the speed at which the dynamic obstacle approaches the lift, ensuring that the obstacle is avoided within a safe distance. When the distance between the lift and a dynamic obstacle approaches the threshold and the obstacle can be avoided in time, the system can adjust the speed of the lift according to the speed of the dynamic obstacle, ensuring the smoothness and continuity of the obstacle avoidance operation and avoiding stopping and waiting. When the distance approaches the threshold and the obstacle cannot be avoided in time, the system can immediately decelerate until it stops completely, avoiding collision with the dynamic obstacle, ensuring safety in any situation.

[0059] When the lift passes a corner, the lift is slowed down and travels in a zigzag or wavy line to ensure that the lift can collect obstacle information on both sides of the corner to the maximum extent possible during its movement, and to ensure that any potential obstacles can be discovered and avoided in time at the corner, thereby improving the immediacy and safety of obstacle avoidance. In addition, the system adjusts the lift's speed and obstacle avoidance path in real time to ensure the smoothness and continuity of the lift's obstacle avoidance operation when passing a corner. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for describing the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention.

[0061] Figure 1 A flowchart of an obstacle avoidance method for a medical transfer machine provided in this application;

[0062] Figure 2 A schematic diagram of the module connection relationship of an obstacle avoidance system of a medical transfer machine provided in this application. DETAILED DESCRIPTION

[0063] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0064] It should be noted that all directional indications in the embodiments of the present application (such as up, down, left, right, front, back, etc.) are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0065] In addition, the descriptions of "first", "second", etc. in this application are only for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in this field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by this application.

[0066] The embodiments of the present application provide an obstacle avoidance method and system for a medical transfer machine, thereby solving the technical problem in the prior art that the transfer machine has poor stability when encountering pedestrians who suddenly change their moving direction or speed.

[0067] like Figure 1 As shown, the technical solution in the embodiment of the present application is to solve the above technical problems, and the overall idea is as follows:

[0068] The present application provides an obstacle avoidance method for a medical transfer machine, comprising the following steps:

[0069] Step S1: Use at least one laser sensor, at least one monocular camera, and at least one depth camera to collect static and dynamic obstacle information in the environment in real time. The laser sensors are deployed at different positions of the lift to detect obstacles in the surrounding environment, especially obstacles at a long distance. The monocular camera is used to capture image information in the environment, which can assist in identifying the shape, color and other visual features of the obstacles. The depth camera provides three-dimensional depth information to help accurately measure the distance and height of the obstacles. Specifically, static obstacle information includes the location and size of the obstacles, dynamic obstacle information includes the location, speed and moving direction of the obstacles, and character status information includes the direction of the feet, the degree of leg bending, facial expressions and line of sight, which helps to judge the character's behavioral intentions.

[0070] Step S2: Using Kalman filtering to process static and dynamic obstacle information from the laser sensor, the monocular camera, and the depth camera to obtain first precise obstacle information;

[0071] Step S3: classifying the first precise obstacle information through a convolutional neural network to obtain precise dynamic obstacle information and precise static obstacle information, continuously tracking the dynamic obstacle, and recording the first dynamic motion trajectory;

[0072] Step S4: Obtain the first dynamic motion trajectory information from step S3, use a recurrent neural network to learn the dynamic obstacle motion pattern, and output a predicted dynamic obstacle trajectory;

[0073] Step S5: Using a static path planning algorithm on a known map, in this embodiment, a Dijkstra algorithm is used, combined with the prediction of dynamic obstacle trajectories to find a preliminary path from the starting point to the target point;

[0074] Step S6: when it is detected that the moving direction or speed of the dynamic obstacle changes to the second precise obstacle information, the second precise obstacle information is immediately used to output and update the dynamic obstacle trajectory;

[0075] Step S7: Output the obstacle avoidance path according to the updated dynamic obstacle trajectory.

[0076] Through multi-sensor real-time collection of static and dynamic obstacle information in the environment, CNN classifies the first precise obstacle information to obtain precise dynamic obstacle information and precise static obstacle information, thereby improving the accuracy of object detection. RNN is used to learn the movement pattern of dynamic obstacles and output the predicted dynamic obstacle trajectory, thereby improving the accuracy of dynamic obstacle prediction. This enables the system to make obstacle avoidance decisions in advance. When a change in the moving direction or speed of a dynamic obstacle is detected, the new obstacle information is immediately used to update the dynamic obstacle trajectory and replan the path to ensure the real-time and effectiveness of obstacle avoidance.

[0077] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0078] Preferably, in step S2, the static and dynamic obstacle information from the laser sensor, the monocular camera and the depth camera is processed using Kalman filtering to obtain the first precise obstacle information, including:

[0079] Step S21: deploy laser sensors, monocular cameras and depth cameras at different positions on the shifter to cover the maximum range of the surrounding environment, collect static and dynamic obstacle information in the environment in real time, determine system state variables, and initialize the covariance matrix P0 to represent the uncertainty of the initial state estimate;

[0080] Step S22: performing time synchronization processing on the information collected by the laser sensor, the monocular camera and the depth camera to ensure that the data from different sensors are synchronized in time, eliminate time deviation, remove noise data, and retain useful information;

[0081] Step S23: state prediction, at each time step, the state transfer equation is used to predict the static and dynamic obstacle information at the current moment according to the static and dynamic obstacle information at the previous moment, and converted into predicted state variables through the observation equation;

[0082] The state transfer equation is:

[0083]

[0084] in, is the state estimate at the current moment, A is the state transfer matrix, B is the control input matrix, u t is the control input, w t is the process noise.

[0085] According to the state transfer equation and the covariance matrix P of the previous moment t , calculate the prediction covariance matrix P t+1 ;

[0086] P t+1 =AP t A T +Q

[0087] where Q is the covariance matrix of the process noise.

[0088] Step S24: Obtain the static and dynamic obstacle information at the current moment, convert it into observation state variables, and update the observation covariance matrix according to the observation state variables;

[0089] The observation equation is:

[0090]

[0091] Where H is the observation matrix, v t is the observation noise.

[0092] According to the observed state variable z t Update the observation covariance matrix R t :

[0093] R t =HP t H T +V

[0094] where V is the covariance matrix of the observation noise.

[0095] Step S25: According to the prediction covariance matrix P t+1 and the observation covariance matrix R t The Kalman gain K is calculated t ;

[0096] K t =P t+1 H T (HP t+1 H T +R) -1

[0097] If the Kalman gain tends to 0, the reliability of the observation value at the current moment is low, and the predicted state variable is selected;

[0098] If the Kalman gain tends to 1, the reliability of the observation value at the current moment is high, and the observed state variable is selected.

[0099] Step S26: According to the Kalman gain K t , predict state variables and the observed state variable z t The updated predicted state variables are obtained, and the updated covariance matrix is ​​obtained. The above steps are repeated at each time step.

[0100] Update the predicted state variables:

[0101]

[0102] Update the covariance matrix:

[0103]

[0104] In a hospital environment, a cart that suddenly appeared in front was detected by a laser sensor. The monocular camera and depth camera provided the shape and distance information of the cart. The Kalman filter integrated this information and predicted the future position of the cart. The lift adjusted the path in time according to the predicted information, successfully avoided the cart and ensured safe passage.

[0105] Preferably, step S6: when it is detected that the moving direction or speed of the dynamic obstacle changes to the second precise obstacle information, immediately using the second precise obstacle information to output and update the dynamic obstacle trajectory, comprises:

[0106] Step S61: The second precise obstacle information is continuously tracked by the convolutional neural network, and the second dynamic motion trajectory is recorded, so as to ensure that the system can update the position information of the dynamic obstacle in real time, and improve the accuracy and real-time performance of obstacle tracking. The CNN can effectively identify and classify obstacles, and ensure that different obstacle types can be accurately distinguished in a dynamic environment. Moreover, through visual feature extraction, a high recognition accuracy rate can be maintained even in the case of occlusion or light changes.

[0107] Step S62: Use a recurrent neural network to learn the second dynamic motion trajectory and output an updated dynamic obstacle trajectory, so that the system can predict the future behavior of the dynamic obstacle based on the historical trajectory data, thereby improving the accuracy and reliability of trajectory prediction. By learning the historical trajectory data and predicting future motion trends, a reliable basis is provided for obstacle avoidance path planning, and it can also help the system better understand the behavior patterns of dynamic obstacles and make obstacle avoidance decisions in advance.

[0108] When the moving direction or speed of a dynamic obstacle changes during the movement of the lift, the system can immediately update the obstacle trajectory and adjust the obstacle avoidance strategy accordingly, thereby improving the system's response speed and obstacle avoidance capabilities, ensuring that correct obstacle avoidance decisions can be made in a dynamic environment in a timely manner to avoid potential collision risks. For example, when a pedestrian suddenly changes direction or speed, the system can immediately adjust the speed or path to ensure safe passage.

[0109] The step S7: outputting an obstacle avoidance path according to the updated dynamic obstacle trajectory, comprises:

[0110] Step S71: Based on the updated information of the dynamic obstacle trajectory, the Dijkstra algorithm is used for path planning, and the path is dynamically adjusted, so that the optimal path from the current position to the target point can be calculated in real time to ensure the real-time and accuracy of the path planning;

[0111] Step S72: Setting a safety distance threshold to ensure that the distance between the lift and the dynamic obstacle is always within a safe range, effectively preventing potential collision risks and ensuring the safety of the lift and pedestrians.

[0112] When the distance between the lift and the dynamic obstacle is close to the threshold and can be avoided in time, the speed of the lift is adjusted according to the speed at which the dynamic obstacle approaches the lift;

[0113] When the distance between the lift and a dynamic obstacle approaches the threshold and the lift cannot avoid the obstacle in time, the lift will decelerate immediately until it stops completely and continue to move forward after the obstacle is cleared.

[0114] In some embodiments, step S73: in step S72, when the distance between the transfer machine and the dynamic obstacle is greater than a threshold, the transfer machine maintains the current speed or accelerates to enter the obstacle avoidance path to improve the passage efficiency;

[0115] When the distance between the transfer machine and the dynamic obstacle is less than or equal to the threshold, the transfer machine is controlled to decelerate, and the deceleration amplitude A of the distance between the transfer machine and the dynamic obstacle is calculated.

[0116] A=k×(Ll)

[0117] Among them, k is the adjustment coefficient, L is the safety distance threshold, and l is the distance between the lift and the dynamic obstacle;

[0118] Calculate the target speed V based on the deceleration amplitude A target ,

[0119] V target =V current -A

[0120] Among them, V current is the current speed of the lift,

[0121] If V target is less than zero, then set V target =0, the lift stops moving;

[0122] Set the speed change rate ΔV. The speed adjustment frequency should be consistent with the update frequency of static and dynamic obstacle information.

[0123] V next -V current -ΔV

[0124] If V next Greater than V target , then set V next =V target ,

[0125] V next is the lower speed of the lift.

[0126] When the distance between the transfer machine and a dynamic obstacle is close to the threshold, the system can dynamically adjust the speed of the transfer machine according to the speed at which the dynamic obstacle approaches the transfer machine to ensure that the obstacle is avoided within a safe distance. When the distance between the transfer machine and a dynamic obstacle is close to the threshold and it can be avoided in time, the system can adjust the speed of the transfer machine according to the speed of the dynamic obstacle to make the speed change of the transfer machine smoother, ensure the smoothness and continuity of the obstacle avoidance operation, and avoid stopping and waiting. When the distance is close to the threshold and it cannot be avoided in time, the system can immediately decelerate until it stops completely to avoid collision with dynamic obstacles, ensuring safety in any situation, especially when the transfer machine is going uphill or downhill, to prevent the speed of the dynamic obstacle from changing too fast and causing side slipping or shaking, so that the transfer machine can pass the dynamic obstacle safely and smoothly.

[0127] In some embodiments, step S6 is further included after step S5. * : Determine the coverage range of the laser sensor, monocular camera and depth camera, adjust the angles of the laser sensor, monocular camera and depth camera according to the specific position of the corner, and ensure that the key areas on both sides of the corner can be covered. When the shift machine approaches the corner, the shift machine is set to be 1 meter away from the corner in this embodiment, and the deceleration amplitude formula is used to gradually reduce the speed of the shift machine to ensure that there is enough reaction time at the corner. The shift machine is controlled to move forward in a zigzag or wavy manner to ensure that the sensor module can collect obstacle information at the corner to the maximum extent;

[0128] Use a monocular camera and a depth camera to collect image information on both sides of the corner, and use a crowd detection algorithm to calculate the density of people on both sides of the corner. The crowd detection algorithm may include but is not limited to background subtraction, motion detection, depth information analysis, etc. Specifically, an image processing algorithm may be used to calculate the number of people on each side, and the driving direction may be selected based on the number of people. After selecting the driving direction, the speed and direction of the lift may be adjusted based on the sensor information to ensure safe passage through the corner.

[0129] According to the detected density of pedestrian flow, the lift is controlled to choose the side with less pedestrian flow as the driving direction, thus reducing the mutual interference with pedestrians;

[0130] If the density of people on both sides of the corner is similar, the lift is controlled to move in the middle to avoid congestion caused by too much people on one side, ensuring that the safest path is selected when there is no obvious difference in the flow of people.

[0131] When the lift passes a corner, the lift is slowed down and travels in a zigzag or wavy line to ensure that the lift can collect obstacle information on both sides of the corner to the maximum extent during travel, and to ensure that any potential obstacles can be discovered and avoided in time at the corner, reducing the bumpy feeling caused by rapid turns. The smooth driving process reduces unnecessary sudden stops or sharp turns, improves the user's riding comfort, and improves the immediacy and safety of obstacle avoidance. In addition, the system adjusts the speed and obstacle avoidance path of the lift in real time, so that the lift can collect obstacle information at the corner to the maximum extent under the premise of straight-line obstacle avoidance driving, so as to plan the obstacle avoidance path in advance and ensure the smoothness and continuity of the obstacle avoidance operation of the lift when passing the corner, so that the medical lift can achieve efficient, safe and intelligent obstacle avoidance functions in complex and changeable environments, ensure the safety of patients and operators, improve user experience and the overall performance of the system, and prevent the lift from affecting the normal walking direction and speed of pedestrians during movement.

[0132] like Figure 2 As shown, a path planning system for a medical transfer machine, applying the method described in any one of the above technical solutions, the system includes:

[0133] A sensing module, including at least one laser sensor, at least one monocular camera and at least one depth camera, for collecting static and dynamic obstacle information in the environment in real time;

[0134] A data processing module is used to process the data information from the sensor module and output the predicted dynamic obstacle trajectory;

[0135] The path planning module is used to plan the path based on the output prediction of dynamic obstacle trajectories;

[0136] A speed control module adjusts the speed of the lift according to the distance between the dynamic obstacle and the lift;

[0137] The corner obstacle avoidance module calculates the density of human traffic through the crowd detection algorithm and switches the movement route of the lift;

[0138] The control system controls the movement of the lift according to the planned path planning information.

[0139] Specifically, the laser sensor and the monocular camera are installed at two ends of the same side of the transfer machine, and the depth camera is installed at the head of the bed of the transfer machine.

[0140] The path planning system of the medical lift collects static and dynamic obstacle information in the environment in real time through laser sensors and monocular cameras installed at both ends of the same side of the lift, and a depth camera installed at the head of the bed; the data processing module uses Kalman filtering to fuse sensor data, and processes information through convolutional neural networks (CNN) and recurrent neural networks (RNN), and outputs predicted dynamic obstacle trajectories; the path planning module uses a static path planning algorithm (such as the Dijkstra algorithm) based on the predicted trajectory to perform preliminary path planning, and replans the path when dynamic obstacles appear; the speed control module adjusts the speed of the lift according to the distance between the dynamic obstacle and the lift to ensure a safe distance; the corner obstacle avoidance module collects image information on both sides of the corner through a monocular camera and a depth camera, and uses a crowd detection algorithm to calculate the density of human traffic, and controls the lift to travel on the side with less human traffic; finally, the control system controls the movement of the lift according to the planned path information to ensure that it can travel safely, smoothly and efficiently in a complex environment.

[0141] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0142] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. An obstacle avoidance method for a medical transfer machine, characterized in that: The following steps are involved: Step S1: collecting static and dynamic obstacle information in the environment in real time through at least one laser sensor, at least one monocular camera and at least one depth camera; Step S2: Using Kalman filtering to process static and dynamic obstacle information from the laser sensor, the monocular camera, and the depth camera to obtain first precise obstacle information; Step S3: classifying the first precise obstacle information through a convolutional neural network to obtain precise dynamic obstacle information and precise static obstacle information, continuously tracking the dynamic obstacle, and recording the first dynamic motion trajectory; Step S4: Obtain the first dynamic motion trajectory information from step S3, use a recurrent neural network to learn the dynamic obstacle motion pattern, and output a predicted dynamic obstacle trajectory; Step S5: Using a static path planning algorithm on a known map, combined with the predicted dynamic obstacle trajectory, to find a preliminary path from the starting point to the target point; Step S6: when it is detected that the moving direction or speed of the dynamic obstacle changes to the second precise obstacle information, the second precise obstacle information is immediately used to output and update the dynamic obstacle trajectory; Step S7: Output the obstacle avoidance path according to the updated dynamic obstacle trajectory.

2. The obstacle avoidance method for a medical lift according to claim 1, characterized in that: In step S2, the static and dynamic obstacle information from the laser sensor, the monocular camera and the depth camera are processed using Kalman filtering to obtain first accurate obstacle information, including: Step S21: deploying laser sensors, monocular cameras and depth cameras at different positions on the shifter to cover the maximum range of the surrounding environment, collecting static and dynamic obstacle information in the environment in real time, determining system state variables, and initializing the covariance matrix; Step S22: performing time synchronization processing on the information collected by the laser sensor, the monocular camera and the depth camera to remove noise data; Step S23: state prediction, at each time step, the state transfer equation is applied to predict the static and dynamic obstacle information at the current moment according to the static and dynamic obstacle information at the previous moment, and converted into predicted state variables through the observation equation; Calculate the predicted covariance matrix based on the state transfer equation and the covariance matrix of the previous moment; Step S24: Obtain the static and dynamic obstacle information at the current moment, convert it into observation state variables, and update the observation covariance matrix according to the observation state variables; Step S25: Calculate the Kalman gain based on the prediction covariance matrix and the observation covariance matrix; Step S26: obtaining updated predicted state variables according to the Kalman gain, predicted state variables and observed state variables, and obtaining an updated covariance matrix. The above steps are repeated at each time step.

3. The obstacle avoidance method for a medical lift according to claim 2, characterized in that: Static and dynamic obstacle information includes the location, size, speed and character status of the obstacle. The character status includes the direction of the foot, the degree of leg bending, facial expression and line of sight direction.

4. The obstacle avoidance method for a medical lift according to claim 2, characterized in that: If the Kalman gain tends to 0, the reliability of the observation value at the current moment is low, and the predicted state variable is selected; If the Kalman gain tends to 1, the reliability of the observation value at the current moment is high, and the observed state variable is selected.

5. The obstacle avoidance method for a medical lift according to claim 1, characterized in that: The step S6: when it is detected that the moving direction or speed of the dynamic obstacle changes to the second precise obstacle information, immediately using the second precise obstacle information to output and update the dynamic obstacle trajectory, comprises: Step S61: continuously tracking the second precise obstacle information through a convolutional neural network and recording a second dynamic motion trajectory; Step S62: Use a recurrent neural network to learn the second dynamic motion trajectory and output an updated dynamic obstacle trajectory.

6. The obstacle avoidance method for a medical lift according to claim 1, characterized in that: The step S7: outputting an obstacle avoidance path according to the updated dynamic obstacle trajectory, comprises: Step S71: Based on the updated dynamic obstacle trajectory information, path planning is performed using the Dijkstra algorithm; Step S72: Setting a safety distance threshold, When the distance between the lift and the dynamic obstacle is close to the threshold and can be avoided in time, the speed of the lift is adjusted according to the speed at which the dynamic obstacle approaches the lift; When the distance between the lift and a dynamic obstacle approaches the threshold and the lift cannot avoid the obstacle in time, the lift will decelerate immediately until it stops completely and continue to move forward after the obstacle is cleared.

7. The obstacle avoidance method for a medical lift according to claim 6, characterized in that: Step S73: In step S72, when the distance between the transfer machine and the dynamic obstacle is greater than the threshold, the transfer machine maintains the current speed or accelerates to enter the obstacle avoidance path; When the distance between the transfer machine and the dynamic obstacle is less than or equal to the threshold, the transfer machine is controlled to decelerate, and the deceleration amplitude A of the distance between the transfer machine and the dynamic obstacle is calculated. A=k×(Ll) Where k is the adjustment coefficient, L is the safety distance threshold, and l is the distance between the lift and the dynamic obstacle; Calculate the target speed V based on the deceleration amplitude A target , In target =V current -A Among them, V current is the current speed of the lift, If V target is less than zero, then set V target =0, the lift stops moving; Set the speed change rate ΔV. The speed adjustment frequency should be consistent with the update frequency of static and dynamic obstacle information. V next -V current -ΔV If V next Greater than V target , then set V next =V target , V next is the lower speed of the lift.

8. The obstacle avoidance method for a medical lift according to claim 1, characterized in that: After step S5, step S6 is also included * : Determine the coverage of the laser sensor, monocular camera and depth camera. When the lift approaches a corner, slow down the lift and control the lift to move forward in a zigzag or wavy manner. Use a monocular camera and a depth camera to collect image information on both sides of the corner, and use a crowd detection algorithm to calculate the density of people on both sides of the corner. According to the detected crowd density, the lift is controlled to choose the side with less crowd as the driving direction. If the density of people on both sides of the corner is similar, the lift is controlled to move in the middle.

9. A path planning system for a medical transfer machine, characterized in that: The method according to any one of claims 1 to 8 is applied, wherein the system comprises: A sensing module, including at least one laser sensor, at least one monocular camera and at least one depth camera, for collecting static and dynamic obstacle information in the environment in real time; A data processing module is used to process the data information from the sensor module and output the predicted dynamic obstacle trajectory; The path planning module is used to plan the path based on the output prediction of dynamic obstacle trajectories; A speed control module adjusts the speed of the lift according to the distance between the dynamic obstacle and the lift; The corner obstacle avoidance module calculates the density of human traffic through the crowd detection algorithm and switches the movement route of the lift; The control system controls the movement of the lift according to the planned path planning information.

10. The path planning system for a medical lift according to claim 9, characterized in that: The laser sensor and the monocular camera are installed at two ends of the same side of the transfer machine, and the depth camera is installed at the head of the bed of the transfer machine.

Images