Real-time data processing and safety protection method and system for embedded disinfection robot

Abstract

The invention discloses a real-time data processing and safety protection method and system for an embedded disinfection robot, and relates to the technical field of self-moving robots, and the method comprises the steps: constructing an environment grid map through a laser radar and an RGB-D camera, and marking static obstacles; by using sensor data and combining with an LSTM model, analyzing obstacle motion characteristics, and detecting and judging a flowing obstacle in real time; a deep learning model of multi-modal data fusion is adopted to assist classification and map updating; according to the distance and the type of the obstacle, starting a hierarchical safety protection mechanism, including dynamic obstacle avoidance, emergency braking and adaptive path planning; the corresponding system comprises an information acquisition mapping module, a flowing obstacle detection module, an obstacle feature fusion module and a safety protection response module. According to the method, the static obstacles and the flowing obstacles are effectively distinguished and processed, and the operation safety and the operation efficiency of the robot in a dynamic environment are improved.

Description

Technical Field

[0001] This invention relates to the field of mobile robot technology, and more specifically, to a method and system for real-time data processing and safety protection of embedded disinfection robots. Background Technology

[0002] Existing disinfection robots mostly adopt an execution mode based on pre-stored maps and fixed paths, which has limited environmental adaptability. In dynamic and complex scenarios such as hospital corridors, outpatient halls, or airport waiting areas, mobile obstacles such as people, vehicles, and temporary piles of objects frequently appear, and traditional solutions are unable to effectively distinguish and respond to them in real time. Most systems only trigger obstacle avoidance through thresholds and cannot determine the attributes of obstacles, resulting in frequent robot starts and stops and path interruptions, which seriously affect the continuity and coverage of disinfection operations. In addition, existing technologies mostly rely on a single sensor for environmental perception and lack a deep fusion mechanism for multi-source information. Point cloud data and image information are often processed independently without achieving collaboration at the feature level. This results in insufficient comprehensive judgment of obstacle contours, textures and motion states. In particular, for partially occluded, lighting-changing or fast-moving objects, the rate of false alarms or missed detections is high. Moreover, the system generally lacks the ability to predict the movement trend of obstacles and cannot plan avoidance strategies in advance, so it can only react passively. In terms of path planning and safety control, traditional methods often use static safety distance thresholds without dynamically adjusting them based on obstacle type, speed, and scene structure. For example, using the same obstacle avoidance parameters in narrow wards and open corridors can easily cause robots to act too conservatively or recklessly. At the same time, existing systems lack adaptive learning mechanisms and cannot optimize recognition models and decision-making strategies based on historical operation data, making it difficult to cope with environmental changes and the emergence of new types of obstacles during long-term operation.

[0003] To address the above problems, this invention proposes a solution. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide a method and system for real-time data processing and safety protection of an embedded disinfection robot, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A method for real-time data processing and safety protection of an embedded disinfection robot, including the steps; Step S1: Scan the work area using LiDAR and RGB-D camera to obtain 3D point cloud data and image information of the environment, perform pose estimation, establish a grid map of the work area, and set the occupancy status and occupancy probability for each grid, marking it as passable, impassable, or unknown. Step S2: Utilize LiDAR, RGB-D camera, and ultrasonic sensors to monitor the surrounding environment in real time and detect moving obstacles. By setting a time window and combining it with the disinfection robot's own motion data, determine whether the obstacle is a moving object. Based on the object's motion state, update the obstacle information in real time and determine whether it will affect the disinfection path. Step S3: Combine RGB images, depth maps, and LiDAR point cloud data to use a multi-channel fusion network for obstacle recognition and classification. Use an attention mechanism to enhance the expression of important features, and use the cross-entropy loss function for classification training. Output obstacle category, location, and confidence. If the obstacle is a moving obstacle, maintain its dynamic state and continue monitoring. If it is a static obstacle, update the map. Step S4: The robot obtains the distance to the obstacle in real time through lidar and other sensors, and calculates the distance difference between the robot and the obstacle. If the distance is less than the set safe distance, the disinfection robot triggers the obstacle avoidance strategy. If the distance to the obstacle is detected to be less than the collision risk threshold, the spraying stops and the safety mode is triggered.

[0006] In a preferred embodiment, step S1 includes the following: The distance and outline information of obstacles are obtained by LiDAR, the image features and depth information of obstacles are obtained by RGB-D camera, and ultrasonic sensors are used for auxiliary distance measurement. Within a set detection time window, obstacles are continuously observed to record their position sequence, and the average speed and moving distance of the obstacles are calculated based on the position sequence. At the same time, the number of observations within the time window is counted. The average speed of the obstacle is calculated based on the difference in its position at different times, and the distance it moves is calculated based on the difference between its initial position and its final position.

[0007] In a preferred embodiment, step S2 includes the following: The working environment is divided into uniform grids, and each grid is assigned an occupancy status and occupancy probability. Occupancy status is used to indicate whether a grid is occupied. The occupancy probability is updated iteratively using a Bayesian occupancy grid model based on the ranging results of the lidar. A misjudgment recovery mechanism is provided. If a grid cell marked as occupied is subsequently observed as empty multiple times, it will be restored to a passable state.

[0008] In a preferred embodiment, step S3 includes the following: Using an LSTM neural network, the time-series data of the obstacle's average speed, travel distance, and number of observations obtained from the aforementioned impact assessment module are used as input to learn the obstacle's motion pattern and predict its future motion trend. A discrimination score threshold is set. The discrimination score is the sum of the ratios of the obstacle's average speed, moving distance, and number of observations to the corresponding preset thresholds. When the score is greater than the set value, the obstacle is determined to be a moving obstacle.

[0009] In a preferred embodiment, step S4 includes the following: A safe distance and a collision distance are set. When the real-time distance between the disinfection robot and an obstacle is less than or equal to the safe distance, a deceleration or obstacle avoidance strategy is activated. When the real-time distance is less than or equal to the collision distance, the disinfection robot is controlled to perform an emergency stop. When the obstacle avoidance strategy is activated, a local path planning algorithm is invoked and combined with the motion prediction information of the obstacles to regenerate the path to bypass the obstacles.

[0010] The embedded disinfection robot real-time data processing and safety protection system includes: an information acquisition and mapping module, a mobile obstacle detection module, an obstacle feature fusion module, and a safety protection response module, with signal connections between each module; Information acquisition and mapping module: Scan the work area with LiDAR and RGB-D camera to obtain 3D point cloud data and image information of the environment, perform attitude estimation, build a grid map of the work area, and set the occupancy status and occupancy probability for each grid, marking it as passable, impassable or unknown. Moving obstacle detection module: Utilizes LiDAR, RGB-D camera, and ultrasonic sensors to monitor the surrounding environment in real time and detect moving obstacles. By setting a time window and combining it with the disinfection robot's own motion data, it determines whether the obstacle is a moving object. Based on the object's motion status, it updates the obstacle information in real time and determines whether it will affect the disinfection path. Obstacle Feature Fusion Module: Combining RGB images, depth maps, and LiDAR point cloud data, a multi-channel fusion network is used to identify and classify obstacles. An attention mechanism is used to enhance the expression of important features, and a cross-entropy loss function is used for classification training. The module outputs the obstacle category, location, and confidence score. If the obstacle is a moving obstacle, its dynamic state is maintained and monitoring continues. If it is a static obstacle, the map is updated. Safety protection response module: It uses lidar and other sensors to obtain the distance to obstacles in real time and calculates the distance difference between itself and the obstacles. If the distance is less than the set safe distance, the disinfection robot triggers the obstacle avoidance strategy. If the distance to the obstacle is detected to be less than the collision risk threshold, the spraying stops and the safety mode is triggered.

[0011] The technical effects and advantages of the embedded disinfection robot real-time data processing and safety protection method of this invention are as follows: This invention significantly improves the accuracy and real-time performance of environmental perception through multi-sensor fusion mapping and dynamic obstacle discrimination. Based on a grid map and Bayesian update mechanism, static obstacles are reliably labeled, reducing real-time computational load; an LSTM model is used to analyze time-series motion data, accurately distinguishing moving obstacles and reducing the false positive rate. Combined with a multimodal deep learning model, intelligent obstacle attribute classification and adaptive map updates are achieved, enhancing the long-term stability of the system. A tiered safety protection mechanism, based on dynamic distance and speed thresholds, enables multi-level responses from deceleration and local path replanning to emergency braking, ensuring human and machine safety. Furthermore, through online learning and model iteration, the system can continuously optimize obstacle recognition and avoidance strategies, improving the efficiency and robustness of disinfection operations in complex scenarios. Attached Figure Description

[0012] Figure 1 This is a schematic diagram illustrating the steps of the embedded disinfection robot real-time data processing and safety protection method of the present invention.

[0013] Figure 2 This is a schematic diagram of the embedded disinfection robot real-time data processing and safety protection system module of the present invention. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] Example This invention discloses a method for real-time data processing and safety protection of an embedded disinfection robot, including the following steps: Step S1: Scan the work area using LiDAR and RGB-D camera to obtain 3D point cloud data and image information of the environment, perform pose estimation, establish a grid map of the work area, and set the occupancy status and occupancy probability for each grid, marking it as passable, impassable, or unknown. Step S2: Utilize LiDAR, RGB-D camera, and ultrasonic sensors to monitor the surrounding environment in real time and detect moving obstacles. By setting a time window and combining it with the disinfection robot's own motion data, determine whether the obstacle is a moving object. Based on the object's motion state, update the obstacle information in real time and determine whether it will affect the disinfection path. Step S3: Combine RGB images, depth maps, and LiDAR point cloud data to use a multi-channel fusion network for obstacle recognition and classification. Use an attention mechanism to enhance the expression of important features, and use the cross-entropy loss function for classification training. Output obstacle category, location, and confidence. If the obstacle is a moving obstacle, maintain its dynamic state and continue monitoring. If it is a static obstacle, update the map. Step S4: The robot obtains the distance to the obstacle in real time through lidar and other sensors, and calculates the distance difference between the robot and the obstacle. If the distance is less than the set safe distance, the disinfection robot triggers the obstacle avoidance strategy. If the distance to the obstacle is detected to be less than the collision risk threshold, the spraying stops and the safety mode is triggered.

[0016] In step S1, the work area is scanned using a LiDAR and an RGB-D camera to acquire 3D point cloud data and image information of the environment. Pose estimation is performed to create a grid map of the work area. Each grid cell is assigned an occupancy status and probability, and is marked as passable, impassable, or unknown. Specific details include: The environmental information of the work area is acquired, including the location of static obstacles such as walls, fixed furniture, and equipment. Simultaneous scanning is performed using LiDAR and RGB-D cameras, and attitude estimation is performed by combining IMU and encoder data. LiDAR provides distance information about the surrounding environment. Path planning requires sensor input based on environmental information. The disinfection robot moves slowly along a predetermined route, scanning the entire work area with LiDAR. The position and angle of the disinfection robot were calculated using odometry based on IMU and encoder data. An RGB-D camera acquires color images and depth maps of the scene to help identify small obstacles and texture information; After acquiring environmental point clouds and images, the disinfection robot divides the work area into uniform grids to facilitate map representation and path planning. To meet the resolution requirements of different scenarios, this embodiment sets the grid size to 0.2 meters × 0.2 meters, which can be adjusted according to the size of the corridor and room. Each grid contains the following information: Occupancy status: The initial status is set to unknown and -1. When the sensor detects that the grid is occupied by an obstacle, its occupancy status is set to 1, meaning it is not passable; if it is detected as empty, it is set to 0, meaning it is passable. Occupancy Probability: To reduce error, a Bayesian occupancy grid model is used, maintaining an occupancy probability value for each grid cell. When the lidar ray passes through the grid, the occupancy probability of the grid is updated based on the reflection ranging result. As the disinfection robot moves and observes multiple times, the occupancy probability of the grid gradually converges. During grid construction, if a grid cell is continuously observed to have obstacles by the sensor, but remains empty in subsequent observations, this may be a misjudgment caused by sensor noise. A misjudgment recovery mechanism needs to be implemented: once a grid cell is marked as occupied, it needs to be continuously... It will only be restored to a passable state if the next observation is empty; To clearly distinguish between static and moving obstacles, the following static objects should be highlighted during the initial sketching phase: Walls and fixtures: Lines or planes are identified using lidar point clouds, and the outlines of walls and rooms are drawn in a grid. Large furniture and equipment, such as nurse stations, operating tables, and instrument cabinets, can be identified by the three-dimensional shape of an RGB-D camera. Restricted areas, such as power distribution rooms and medicine warehouses, should be marked as no-entry zones based on the building floor plan. Disinfection robot docking area: An area used for replenishing disinfectant or charging disinfection robots, marked as a service station grid; The above information is entered by technicians or confirmed on-site through manual surveying before the disinfection robot's first operation, and marked as static obstacles as the static part of the initial map; After the static obstacle is marked, the disinfection robot does not need to continuously detect it when performing the disinfection task. It only corrects the detection results in steps S2 and S3. This reduces the real-time calculation burden and avoids the frequent inclusion of static obstacles in dynamic judgment. With a reasonable grid size and occupancy probability model, the disinfection robot can achieve accurate positioning and navigation through sensors without the help of an external positioning system, creating conditions for subsequent algorithms.

[0017] In step S2, the surrounding environment is monitored in real time using LiDAR, an RGB-D camera, and an ultrasonic sensor to detect moving obstacles. By setting a time window and combining it with the disinfection robot's own motion data, it is determined whether the obstacle is a moving object. Based on the object's motion state, the obstacle information is updated in real time, and it is determined whether it will affect the disinfection path. Specific details include: Mobile obstacles refer to non-fixed objects, such as patients' suitcases, mobile beds, trolleys, or people who stay briefly in the hospital. These obstacles will appear and move temporarily in the disinfection robot's working area, affecting the disinfection path planning. In order to avoid frequent map updates or misjudgments, a real-time detection and judgment algorithm was designed based on the grid map, which is divided into parameter acquisition, judgment rules, periodic checks, and processing strategies. The robot uses 360° distance data provided by LiDAR to detect the outlines of surrounding objects; it acquires image features and depth information of objects through an RGB-D camera; and it uses an IMU and encoder to provide the motion state of the disinfection robot itself and perform coordinate transformation. Since the depth camera can provide dense neighborhood features and the LiDAR has high distance accuracy, the combination of the two can improve the detection accuracy. Time window settings: Set a detection time window Within this window, the location of each suspected obstacle is continuously observed and recorded. and time The robot determines the motion state of obstacles by calculating their displacement and velocity within a window. The length of this window needs to be adjusted based on the robot's speed and the complexity of the working environment: if the environment has high pedestrian traffic, the window can be shortened to speed up detection; if the environment is relatively still, the window can be lengthened to reduce false positives. To achieve accurate obstacle detection, multiple sensors are used to acquire environmental data. The main sensors include LiDAR, RGB-D cameras, and ultrasonic sensors. These sensors acquire information such as the speed, position, and distance of obstacles in real time, which serves as input data for the LSTM model. The disinfection robot system calculates the average speed of obstacles in real time using lidar, RGB-D cameras, and ultrasonic sensors, assuming that at time... Location information of obstacles collected from sensors Then the average speed of the obstacle It can be calculated using the following formula: ;in, For the first The location information of the obstacle at any given time. The number of moments.

[0018] The system calculates the distance an obstacle moves within a set time period using data from LiDAR and RGB-D cameras. Assuming the initial position of the obstacle is The final position is The distance the obstacle moves The calculation is: L = ; Number of observations of obstacles This refers to the frequency with which the disinfection robot's sensors detect obstacles within a specific time period. By setting a fixed time window, the number of observations within that time period is calculated, noting whether an obstacle is within the designated area each time. Assuming that within the time window... If an obstacle is successfully detected in each of the several observations, then: ; Long Short-Term Memory (LSTM) networks are deep learning models used to process time-series data. They are able to capture long-term dependencies in the data. To identify flowing obstacles, the LSTM model takes the average velocity of the obstacle as input. Distance of movement and number of observations The time-series data is used to train the model, allowing it to learn the historical movement patterns of obstacles and predict their future movement trends. The system can learn from the historical motion data of obstacles and capture their motion patterns through the LSTM model. The LSTM network structure consists of multiple hidden layers, each with multiple memory units, which can store and update the motion information of obstacles in the time dimension. The output of the LSTM model includes the future position, velocity, and direction of motion of the obstacle. By analyzing this output data, it is possible to accurately determine whether the obstacle is a moving obstacle. Specifically, the LSTM model predicts the motion state of obstacles in the future based on the current input data. If the predicted motion of the obstacle will affect the path of the disinfection robot, or if the motion of the obstacle is unstable, it is determined to be a moving obstacle. Conversely, if the motion of the obstacle is stable and does not affect the path of the disinfection robot, it is determined to be a static obstacle. To identify flowing obstacles, specific thresholds and criteria were set, and the results were obtained by analyzing the output of the LSTM model and the average velocity acquired in real time. Distance of movement and number of observations Based on the time-series data, determine whether the obstacle is a moving obstacle and set corresponding thresholds, namely the velocity threshold. Distance threshold and the number of times threshold The discrimination score is set as: S = The judgment criteria are as follows: like If so, the obstacle is determined to be a moving obstacle; like If the condition is met, then the obstacle is determined to be a static obstacle.

[0019] To further improve the accuracy of moving obstacle detection, multimodal data fusion technology is combined, integrating data from multiple sensors such as LiDAR, RGB-D camera, and ultrasonic sensor. Through deep learning model optimization, it can achieve more accurate moving obstacle judgment based on obstacle information provided by different sensors.

[0020] In step S3, a multi-channel fusion network is used to identify and classify obstacles by combining RGB images, depth maps, and LiDAR point cloud data. An attention mechanism is used to enhance the expression of important features, and a cross-entropy loss function is used for classification training. The network outputs the obstacle category, location, and confidence score. If the obstacle is a moving obstacle, its dynamic state is maintained and monitoring continues; if it is a static obstacle, the map is updated. Specific details include: Collect image and point cloud data in typical scenarios, including various moving and static obstacles. Each sample contains RGB image, depth map and LiDAR point cloud, including samples under different angles, lighting conditions and occlusion conditions. The input can include RGB images and depth information from LiDAR. Since the structure of LiDAR point cloud data is different from that of images, it needs to be projected onto a two-dimensional plane aligned with the camera and represented using a depth map. A multi-channel fusion network can be used, with the first branch processing the RGB image, the second branch processing the depth map, and the third branch processing the clustered LiDAR reflection intensity map. After each branch extracts features, they are fused in the middle layer. An attention mechanism is used to enhance feature expression and improve recognition performance. Classification training can be performed using the cross-entropy loss function, with the Adam optimizer employed and appropriate learning rate and batch size set. Feature pyramids and multi-scale prediction heads can be introduced into the network structure. Binary classification labels for flowing and static obstacles are added to the output layer of the model. The velocity estimate obtained by optical flow information or Kalman filtering in historical frames is used as an auxiliary feature, and the motion state of the target in each sample is used as a label. The model learns the mapping relationship between image features and motion features to predict dynamic attributes. After an object is marked as a mobile obstacle, its corresponding RGB image, depth map, and local point cloud data from LiDAR are input into a deep learning model for inference. The output includes category, location, confidence level, and dynamic attributes. The inference results are fed back to the obstacle management module through a message system.

[0021] Confidence threshold setting: To avoid false positives, a confidence threshold can be set. Only when the confidence level of the category or dynamic attribute identified by the model is higher than Only when the threshold is met is the judgment result considered reliable; otherwise, the number of observations needs to be increased or manual confirmation is required. This threshold can be dynamically adjusted based on experimental results. After the deep learning model provides a classification result for a suspected obstacle, it determines whether to update the grid map based on the category and dynamic attributes: If the model determines that an object is a long-term occupation item and its dynamic attribute is static, the grid containing that object is marked as "impassable" and updated in the static map. This means that subsequent path planning will permanently avoid that grid until the map is updated again. The update process records the update time and object category for staff to review.

[0022] If the model confirms that the object is still in motion or is a temporary stop, the static map is not updated; instead, the object's information is maintained in the dynamic obstacle list for real-time path planning and avoidance. This meets the requirements of dynamic obstacle scenarios, which avoid frequent map modifications and instead solve the problem through real-time avoidance strategies. When the distance between the disinfection robot and the obstacle is less than the defined safe distance, the collision avoidance mechanism needs to be activated and prioritized, demonstrating the importance of dynamic avoidance. To avoid misjudgments and omissions, the system is designed with a feedback mechanism: when the model confidence is low or the result is uncertain, the captured image can be uploaded to the backend for remote operation and maintenance personnel to confirm. If the backend confirms that the object is a fixed obstacle, the map can be updated via remote command; if it is confirmed to be a dynamic obstacle, real-time monitoring continues. When the static map is updated, the path planning module needs to recalculate the optimal route. At the global level, it uses A / D algorithms to search for paths to the target; at the local level, it uses dynamic windowing or artificial potential field methods for real-time obstacle avoidance. Related research indicates that in the path planning of mobile disinfection robots, safety is achieved by increasing repulsive force when the obstacle is at a safe distance; therefore, a safe distance parameter needs to be considered in path planning. This involves adding a cost inversely proportional to the distance to obstacles to the planning cost function; As the environment or disinfected items are updated, the model needs to continuously learn new categories or scene changes. Therefore, an adaptive update strategy for the model was designed: During operation, images and sensor data of misclassified or new categories of objects are archived along with manually labeled information, and these data are regularly organized to form an incremental training set; By leveraging online learning technology, the model can be incrementally trained at regular intervals, incorporating new data, without needing to train the entire model from scratch. Transfer learning methods can be used to fine-tune the existing weights, enabling rapid adaptation to new obstacle types or scene changes. To ensure security, offline testing is performed after each model update to verify its accuracy and efficiency. If the new model performs poorly, it can be rolled back to a previous version. The version management system records the time of each update, the amount of training data, and performance metrics for easy traceability. By using deep learning to assist in judgment, we can not only improve the accuracy of identifying moving obstacles, but also achieve intelligent updates of the grid map, ensuring the system's adaptability and long-term stable operation. In step S4, the distance to obstacles is acquired in real time using lidar and other sensors, and the distance difference between the robot and obstacles is calculated. If the distance is less than the set safe distance, the disinfection robot triggers an obstacle avoidance strategy. If the distance to an obstacle is detected to be less than a collision risk threshold, spraying stops and a safety mode is triggered. Specific details include: Safety protection is a crucial part of this embodiment. Its core is to ensure that the disinfection robot does not pose a danger to personnel and equipment when performing disinfection tasks in dynamic environments, while also ensuring its own safety. The safety protection mechanism combines the detection and identification results of the aforementioned three steps to set the starting conditions and specific processing procedures.

[0023] The disinfection robot experiences a certain repulsive force when maintaining a safe distance from obstacles, thus ensuring driving safety. This safe distance, for example, 1.0 meter, is set based on the robot's operating speed and braking distance. During path planning, this safe distance is considered when the disinfection robot is at a safe distance from obstacles. satisfy When this happens, the obstacle avoidance strategy is activated; Collision criterion: If the distance between the disinfection robot and the obstacle... Less than the collision distance If a collision risk is detected, spraying must be stopped immediately and emergency braking must be initiated. When the distance is less than the safe distance, the collision avoidance mechanism must be activated and given the highest control priority. When a dynamic obstacle is detected ahead, the speed If the speed exceeds the preset threshold and its direction of motion intersects with the path of the disinfection robot, a dynamic collision may occur, and the robot should slow down or stop in advance. It is equipped with an emergency stop button and a communication monitoring program. When the emergency stop button is triggered or the communication between the disinfection robot and the control center is lost, it will immediately enter the safety mode. When the disinfection robot approaches a safe distance At this time, the controller issues a deceleration command and performs linear or nonlinear deceleration control based on the distance difference, such as using a proportional control strategy. ;when hour, 0; If the obstacle is a dynamic object and the predicted path intersects, the local path planning algorithm is called to regenerate the path, and the driving direction and angular velocity of the disinfection robot are adjusted to make it bypass the obstacle. During the bypassing process, the safe bypass radius is calculated using motion prediction information from Kalman filtering or deep learning models. When the emergency stop button is pressed or a communication failure occurs, the motor power is immediately disconnected, the braking mechanism is activated, the disinfection robot stops, and all actuators enter a safety lock state.

[0024] After an emergency stop, the disinfection robot will perform a self-check procedure, including checking the sensor status, battery status, and communication link. Once the fault is resolved and a restart command is received, the protection mode can be deactivated. To improve safety, this embodiment employs redundant ranging sensors and algorithms. For example, ultrasonic and infrared sensors are used for distance measurement when the lidar fails. When the Kalman filter state is unreliable, multiple filters are used in conjunction with a deep learning model to estimate the obstacle motion. Different locations are set up and Values, such as the small space in the ward, The spraying distance can be set to 0.8 meters. If the hospital corridor is spacious, it can be set to 1.2 meters. In addition, the spraying dosage and speed also need to be adjusted according to the area to reduce interference with people. In the future, reinforcement learning technology can be combined to train a safety policy network using historical operation data, allowing disinfection robots to automatically adjust their action strategies according to complex environments. For example, in the SAC-LSTM research, by introducing memory units, the agent can make more reasonable decisions based on history and current state, and predict the future position of dynamic obstacles. If the disinfection robot experiences a minor collision during operation, the system identifies the collision event through an accelerometer or abnormal motor current, records the collision location, time, and cause, and uploads it to the monitoring platform for analysis. When the disinfection robot is unable to continue operating due to complex environment or sensor malfunction, the system sends a signal requesting manual intervention, waiting for staff to remove obstacles or perform maintenance on the disinfection robot. At this time, the disinfection robot stops in a safe area and waits. When sensor data shows abnormal fluctuations, operations will be suspended and a system check will be performed. If the problem cannot be resolved in a short time, maintenance personnel will be notified for handling.

[0025] This invention discloses an embedded disinfection robot real-time data processing and safety protection system, including: an information acquisition and mapping module, a mobile obstacle detection module, an obstacle feature fusion module, and a safety protection response module, with signal connections between the modules; Information acquisition and mapping module: Scan the work area with LiDAR and RGB-D camera to obtain 3D point cloud data and image information of the environment, perform attitude estimation, build a grid map of the work area, and set the occupancy status and occupancy probability for each grid, marking it as passable, impassable or unknown. Moving obstacle detection module: Utilizes LiDAR, RGB-D camera, and ultrasonic sensors to monitor the surrounding environment in real time and detect moving obstacles. By setting a time window and combining it with the disinfection robot's own motion data, it determines whether the obstacle is a moving object. Based on the object's motion status, it updates the obstacle information in real time and determines whether it will affect the disinfection path. Obstacle Feature Fusion Module: Combining RGB images, depth maps, and LiDAR point cloud data, a multi-channel fusion network is used to identify and classify obstacles. An attention mechanism is used to enhance the expression of important features, and a cross-entropy loss function is used for classification training. The module outputs the obstacle category, location, and confidence score. If the obstacle is a moving obstacle, its dynamic state is maintained and monitoring continues. If it is a static obstacle, the map is updated. Safety protection response module: It uses lidar and other sensors to obtain the distance to obstacles in real time and calculates the distance difference between itself and the obstacles. If the distance is less than the set safe distance, the disinfection robot triggers the obstacle avoidance strategy. If the distance to the obstacle is detected to be less than the collision risk threshold, the spraying stops and the safety mode is triggered.

[0026] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0027] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0028] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and inventive constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0029] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0030] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0031] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for real-time data processing and safety protection of an embedded disinfection robot, characterized in that, Includes steps; Step S1: Scan the work area using LiDAR and RGB-D camera to obtain 3D point cloud data and image information of the environment, perform pose estimation, establish a grid map of the work area, and set the occupancy status and occupancy probability for each grid, marking it as passable, impassable, or unknown. Step S2: Utilize LiDAR, RGB-D camera, and ultrasonic sensors to monitor the surrounding environment in real time and detect moving obstacles. By setting a time window and combining it with the disinfection robot's own motion data, determine whether the obstacle is a moving object. Based on the object's motion state, update the obstacle information in real time and determine whether it will affect the disinfection path. Step S3: Combine RGB images, depth maps, and LiDAR point cloud data to use a multi-channel fusion network for obstacle recognition and classification. Use an attention mechanism to enhance the expression of important features, and use the cross-entropy loss function for classification training. Output obstacle category, location, and confidence. If the obstacle is a moving obstacle, maintain its dynamic state and continue monitoring. If it is a static obstacle, update the map. Step S4: The robot obtains the distance to the obstacle in real time through lidar and other sensors, and calculates the distance difference between the robot and the obstacle. If the distance is less than the set safe distance, the disinfection robot triggers the obstacle avoidance strategy. If the distance to the obstacle is detected to be less than the collision risk threshold, the spraying stops and the safety mode is triggered.

2. The method for real-time data processing and safety protection of the embedded disinfection robot according to claim 1, characterized in that, The distance and contour information of obstacles are obtained by LiDAR, the image features and depth information of obstacles are obtained by RGB-D camera, and the distance is assisted by ultrasonic sensor.

3. The method for real-time data processing and safety protection of the embedded disinfection robot according to claim 2, characterized in that, Within a set detection time window, obstacles are continuously observed to record their position sequence, and the average speed and distance traveled by the obstacles are calculated based on the position sequence. At the same time, the number of observations within the time window is counted.

4. The method for real-time data processing and safety protection of the embedded disinfection robot according to claim 3, characterized in that, The average speed of the obstacle is calculated based on the difference in its position at different times, and the distance it moves is calculated based on the difference between its initial position and its final position.

5. The method for real-time data processing and safety protection of the embedded disinfection robot according to claim 1, characterized in that, The working environment is divided into uniform grids, and each grid is assigned an occupancy status and occupancy probability. Occupancy status indicates whether a grid is occupied. The occupancy probability is updated iteratively using a Bayesian occupancy grid model based on the ranging results from the LiDAR.

6. The method for real-time data processing and safety protection of the embedded disinfection robot according to claim 5, characterized in that, A misjudgment recovery mechanism is provided. If a grid cell marked as occupied is subsequently observed as empty multiple times, it will be restored to a passable state.

7. The method for real-time data processing and safety protection of the embedded disinfection robot according to claim 1, characterized in that, Using an LSTM neural network, the time-series data of the obstacle's average speed, moving distance, and number of observations obtained from the aforementioned impact assessment module are used as input to learn the obstacle's motion pattern and predict its future motion trend.

8. The method for real-time data processing and safety protection of the embedded disinfection robot according to claim 7, characterized in that, A discrimination score threshold is set. The discrimination score is the sum of the ratios of the obstacle's average speed, moving distance, and number of observations to the corresponding preset thresholds. When the score is greater than the set value, the obstacle is determined to be a moving obstacle.

9. The method for real-time data processing and safety protection of the embedded disinfection robot according to claim 1, characterized in that, A safe distance and a collision distance are set. When the real-time distance between the disinfection robot and an obstacle is less than or equal to the safe distance, a deceleration or obstacle avoidance strategy is activated. When the real-time distance is less than or equal to the collision distance, the disinfection robot is controlled to perform an emergency stop. When the obstacle avoidance strategy is activated, a local path planning algorithm is invoked and combined with the motion prediction information of the obstacles to regenerate the path to bypass the obstacles.

10. An embedded disinfection robot real-time data processing and safety protection system, used to implement the embedded disinfection robot real-time data processing and safety protection method according to any one of claims 1-9, characterized in that: Information acquisition and mapping module: Scan the work area with LiDAR and RGB-D camera to obtain 3D point cloud data and image information of the environment, perform attitude estimation, build a grid map of the work area, and set the occupancy status and occupancy probability for each grid, marking it as passable, impassable or unknown. Moving obstacle detection module: Utilizes LiDAR, RGB-D camera, and ultrasonic sensors to monitor the surrounding environment in real time and detect moving obstacles. By setting a time window and combining it with the disinfection robot's own motion data, it determines whether the obstacle is a moving object. Based on the object's motion status, it updates the obstacle information in real time and determines whether it will affect the disinfection path. Obstacle Feature Fusion Module: Combining RGB images, depth maps, and LiDAR point cloud data, a multi-channel fusion network is used to identify and classify obstacles. An attention mechanism is used to enhance the expression of important features, and a cross-entropy loss function is used for classification training. The module outputs the obstacle category, location, and confidence score. If the obstacle is a moving obstacle, its dynamic state is maintained and monitoring continues. If it is a static obstacle, the map is updated. Safety protection response module: It uses lidar and other sensors to obtain the distance to obstacles in real time and calculates the distance difference between itself and the obstacles. If the distance is less than the set safe distance, the disinfection robot triggers the obstacle avoidance strategy. If the distance to the obstacle is detected to be less than the collision risk threshold, the spraying stops and the safety mode is triggered.

Images