A laser obstacle avoidance system for mobile devices

Abstract

The present invention relates to the technical field of laser obstacle avoidance for mobile devices and discloses a laser obstacle avoidance system for mobile devices, comprising a multi-line laser scanning module for emitting multiple parallel laser beams and receiving reflected light, a data processing module for processing collected three-dimensional information and generating a preliminary obstacle avoidance strategy, a scene recognition module for real-time identification of the current application scene and outputting a classification of the current scene information, a strategy switching unit for receiving current scene information and dynamically outputting an obstacle avoidance strategy for the current scene, and an actuator for controlling the movement of the mobile device according to the obstacle avoidance strategy, thereby enabling the mobile device to intelligently respond to obstacles in various scenes and improving the analysis accuracy and response speed of the mobile device in responding to forest roads and mountain roads.

Description

Technical Field

[0001] The invention relates to the technical field of laser obstacle avoidance for mobile devices and discloses a laser obstacle avoidance system for mobile devices. Background Art

[0002] Existing obstacle avoidance technologies for mobile devices often rely on single-line laser scanning or simple ultrasonic ranging methods. These methods suffer from significant deficiencies in accuracy, real-time performance, and environmental adaptability. Especially in complex scenarios like forest and mountain trails, existing technologies often struggle to accurately identify obstacles, posing a serious risk of collision for mobile devices. Summary of the Invention

[0003] The purpose of this section is to summarize some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the abstract and title of this application to avoid obscuring the purpose of this section, the abstract and the title of the invention, and such simplifications or omissions should not be used to limit the scope of the present invention.

[0004] To solve the above technical problems, the main purpose of the present invention is to provide a laser obstacle avoidance system for a mobile device, wherein the laser obstacle avoidance system for a mobile device comprises:

[0005] Multi-line laser scanning emits multiple parallel laser beams and receives reflected light to obtain three-dimensional information, processes the three-dimensional information, and generates a preliminary obstacle avoidance strategy;

[0006] The scene recognition module receives the processed three-dimensional information in real time, identifies the current application scene, and outputs a classification result of the current scene information;

[0007] The strategy switching unit receives the classification result of the current scene information and dynamically outputs the obstacle avoidance strategy of the current scene;

[0008] The actuator controls the movement of the mobile device according to the obstacle avoidance strategy.

[0009] As a preferred solution of the laser obstacle avoidance system for a mobile device of the present invention, wherein:

[0010] The multi-line laser scanning comprises:

[0011] The laser transmitter is used to emit multiple parallel laser beams spaced at a preset distance;

[0012] The laser receiver is used to receive the laser beam emitted by the laser transmitter and reflected by the surrounding environment, and convert the received laser beam into an electrical signal;

[0013] The scanning mechanism is used to drive the laser emitter and the laser receiver to rotate or translate within a preset range;

[0014] The data processing module is used to process the electrical signal converted by the laser receiver to generate data representing the three-dimensional structure of the surrounding environment.

[0015] As a preferred solution of the laser obstacle avoidance system for a mobile device of the present invention, wherein:

[0016] Inputting the processed three-dimensional coordinates of the three-dimensional information into a three-dimensional space rigid transformation field, wherein the three-dimensional space rigid transformation field constructs a rigid motion pattern of a neighborhood of an object on a forest road or a mountain road;

[0017] Extracting eigenvalues ​​of the rigid motion mode of the neighborhood, encoding the eigenvalues ​​into a spatial transformation matrix, evaluating the accuracy of the spatial transformation matrix, sorting by the accuracy, and retaining stable features of the current application scene acquired by the laser sensor;

[0018] The scene recognition module is used to classify different environmental scenes and output current scene information.

[0019] As a preferred solution of the laser obstacle avoidance system for a mobile device of the present invention, wherein:

[0020] The spatial transformation matrix is ​​transformed into three-dimensional convolution kernel deformation parameters. Differentiated kernel deformation calculation is performed on the X, Y, and Z axes of the three-dimensional spatial rigid transformation field through a dynamic convolution kernel. The differentiated kernel deformation calculation includes vibration-sensitive axis feature processing and stable axis feature processing. Smoothing filtering is applied to the vibration-sensitive axis, and feature extraction is performed on the stable axis.

[0021] If the differential kernel deformation anomaly index exceeds the standard threshold, the dynamic convolution kernel is frozen, the baseline feature calculation is switched, and the pre-stored scene template is called for calibration.

[0022] As a preferred solution of the laser obstacle avoidance system for a mobile device of the present invention, wherein:

[0023] The current scene information includes forest roads and mountain roads;

[0024] By receiving the current scene information, dynamically analyzing the current scene information, and outputting the analysis results to the strategy switching unit, the strategy switching unit dynamically outputs the forest road obstacle avoidance strategy and the mountain road obstacle avoidance strategy.

[0025] As a preferred solution of the laser obstacle avoidance system for a mobile device of the present invention, wherein:

[0026] Dynamic analysis of forest paths through vegetation density analysis and individual plant growth;

[0027] Use laser transmitters to obtain 3D point cloud data of the mobile device environment and analyze the spatial distribution of vegetation;

[0028] Use radius filtering to remove noise points and separate ground points from vegetation points;

[0029] The 3D point cloud data is flattened and divided into multiple grids, and the vegetation density corresponding to the number of point clouds in each grid is obtained. The vegetation density levels are divided according to the vegetation density, which are low density, passable, medium density, cautious pass, and high density, no pass.

[0030] Further, analyzing individual obstacles in the vegetation density, and identifying the height, width, and volume of the individual obstacles based on the acquired 3D point cloud data of the mobile device environment;

[0031] If the width and volume are greater than the maximum width that a mobile device can pass through, the mobile device will select a forced avoidance strategy. Low density can be prohibited by the strategy. If the width and volume are not greater than the maximum width that a mobile device can pass through, the mobile device will select a free avoidance strategy.

[0032] As a preferred solution of the laser obstacle avoidance system for a mobile device of the present invention, wherein:

[0033] The mountain road recognition analysis uses laser radar to obtain three-dimensional point cloud data of the terrain surface, removes abnormal points through radius filtering, and separates ground points from non-ground points through the RANSAC algorithm;

[0034] After data preprocessing is completed, the slope and curvature characteristics of the terrain are extracted, the steepness of the local terrain is analyzed, and the slope of each point cloud data point and its neighboring points is obtained;

[0035] The method for obtaining the slope of each point cloud data point and its adjacent points includes:

[0036] Calculate the elevation difference and horizontal distance between the first point cloud data point and each adjacent point;

[0037] Obtaining the slope of each neighboring point of the first point cloud data point by calculating the inverse tangent trigonometric function of the elevation difference and horizontal distance between the first point cloud data point and each neighboring point;

[0038] Taking the average of the slopes of all neighboring points of the first point cloud data point as the slope value of the first point cloud data point;

[0039] The curvature of the first point cloud data point is calculated by the quotient of the second derivative of the slope and the slope value of the first point data point.

[0040] As a preferred solution of the laser obstacle avoidance system for a mobile device of the present invention, wherein:

[0041] Analyzing the steepness of the local terrain of the mobile device by obtaining the slope and curvature;

[0042] Analyzing the local terrain of the mobile device includes:

[0043] Divide the terrain into multiple local areas, segment the local areas by setting multiple windows, and calculate the average slope and curvature in each window;

[0044] Starting from the current position of the mobile device, a prediction point is set at a preset distance along the direction of movement of the mobile device. The slope of each prediction point is calculated based on the current slope and curvature. If the curvature is positive, the slope gradually increases, and if the curvature is negative, the slope gradually decreases.

[0045] If the terrain is bumpy, the window size is increased; if the terrain is flat, the window size is reduced. There are multiple collection points in the local area. If the slope of a collection point exceeds the set threshold with the slope of the previous prediction point, local resampling is started to recalculate the slope of the current mobile device location.

[0046] As a preferred solution of the laser obstacle avoidance system for a mobile device of the present invention, wherein:

[0047] The mountain road identification analysis also includes identifying rocks and predicting rock falls;

[0048] Scan rock data using a laser transmitter and obtain rock geometric features, including height, width, and volume, through rock clustering. Rocks and rockfalls are detected using a rock target recognition unit. The rock and rockfall recognition results are combined with the 3D point cloud data to analyze the mountain road and determine the location, size, and motion state of the rocks and rockfalls.

[0049] The location, size and movement status of the rocks and falling rocks are used to plan the path of the mobile equipment and generate early warning information.

[0050] As a preferred solution of the laser obstacle avoidance system for a mobile device of the present invention, wherein:

[0051] The mobile device includes a small-diameter gear and a large-diameter gear. The roll angle of the mobile device is obtained by predicting the slope of the point. If the roll angle of the mobile device does not exceed the maximum load threshold of the device, the mobile device switches to the small-diameter gear output power and adjusts the motor output torque to control the rotation of the small-diameter gear. The output power of the small-diameter gear drives the large-diameter gear to rotate through the transmission device, thereby controlling the speed of the mobile device over the slope.

[0052] Beneficial effects of the present invention:

[0053] The system can identify and analyze the different characteristics of forest roads and mountain roads and generate targeted obstacle avoidance strategies. The scene recognition module can identify the current environment type in real time and dynamically switch obstacle avoidance strategies to ensure the adaptability of the equipment in different scenarios and adjust the obstacle avoidance strategy in real time. The system can dynamically adjust the path planning based on the real-time analysis results to ensure the safe navigation of the equipment in complex environments. The system adopts a modular design, and each module is independent of each other and works together, which is also convenient for single module maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort. Among them:

[0055] Figure 1 This is a diagram showing the composition of a laser obstacle avoidance system for a mobile device according to the present invention;

[0056] Figure 2 A schematic diagram of a slope obstacle avoidance system for a mobile device according to the present invention;

[0057] Figure 3 A schematic diagram of the bumpy road coordinates of a laser obstacle avoidance system for a mobile device of the present invention;

[0058] Figure 4 This is a schematic diagram of a laser obstacle avoidance system for a mobile device of the present invention processing collisions between branches of a mobile device while traveling through a forest. DETAILED DESCRIPTION

[0059] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0060] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0061] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0062] Example 1:

[0063] like Figure 1 As shown, a laser obstacle avoidance system for a mobile device includes:

[0064] Multi-line laser scanning emits multiple parallel laser beams and receives reflected light to obtain three-dimensional information, processes the three-dimensional information, and generates a preliminary obstacle avoidance strategy;

[0065] The multi-line laser scanning comprises:

[0066] The laser transmitter is used to emit multiple parallel laser beams spaced at a preset distance;

[0067] The laser receiver is used to receive the laser beam emitted by the laser transmitter and reflected by the surrounding environment, and convert the received laser beam into an electrical signal;

[0068] The scanning mechanism is used to drive the laser emitter and / or the laser receiver to rotate or translate within a preset range;

[0069] The data processing module is used to process the electrical signal converted by the laser receiver to generate data representing the three-dimensional structure of the surrounding environment.

[0070] The preliminary obstacle avoidance strategy includes identifying road boundaries and avoiding obstacles.

[0071] The scene recognition module receives the processed three-dimensional information in real time, identifies the current application scene, and outputs a classification result of the current scene information;

[0072] Inputting the processed three-dimensional coordinates of the three-dimensional information into a three-dimensional space rigid transformation field, wherein the three-dimensional space rigid transformation field constructs a rigid motion pattern of a neighborhood of an object on a forest road or a mountain road;

[0073] Specifically, a specific implementation method for establishing a three-dimensional rigid transformation field includes:

[0074] Receive the three-dimensional point cloud data collected by the lidar in real time, first perform noise reduction processing, merge adjacent duplicate points, simplify the data volume, divide the scene into multiple local neighborhood units, and distinguish different objects.

[0075] A dynamic weighted registration algorithm is used to compare the point cloud changes at adjacent time points, assigning high weights to static objects and low weights to dynamic objects to reduce interference. At the same time, random sampling is used to screen out the optimal rigid motion model for each neighborhood unit to eliminate the influence of noise and outliers.

[0076] The specific rigid motion model includes the rotation and translation parameters of the mobile device.

[0077] The rigid motion parameters of each neighborhood are mapped to a three-dimensional spatial grid to form a rigid transformation field covering the entire scene. The transformation parameters are smoothed using temporal filtering technology to eliminate instantaneous noise caused by sensor jitter and maintain motion continuity.

[0078] Extracting eigenvalues ​​of rigid motion modes of a neighborhood, encoding the eigenvalues ​​into a spatial transformation matrix, evaluating the accuracy of the spatial transformation matrix, and sorting by the accuracy to retain stable features of the current application scenario acquired by the laser sensor;

[0079] Specifically, a specific implementation method for extracting characteristic values ​​of rigid motion modes includes:

[0080] For the point cloud data in each neighborhood, its three-dimensional spatial distribution characteristics are statistically analyzed, and the main motion direction and intensity of the point cloud are calculated. Through main direction analysis, the dynamic area of ​​the branches swaying in the wind and the stable area of ​​the fixed road surface are distinguished, and key indicators such as motion parameters and direction angles are extracted.

[0081] The offset angles and intensity displacements of the X, Y, and Z axes of motion are encoded into multidimensional feature vectors. The reliability of each feature vector is evaluated through a statistical model, the stable parameters of static objects are retained, and noise or temporary interference data are eliminated.

[0082] like Figure 3 As shown, assuming that the mobile device is traveling on a horizontal road (X-axis), the width of the horizontal road is represented by the Y-axis, and the vibration of the mobile device caused by the bumpy road is represented by the Z-axis vibration.

[0083] A method for establishing a statistical model includes:

[0084] In a continuous time window, the multidimensional feature vector of each neighborhood unit is recorded, including the X, Y, and Z axis offset angles and displacement intensity. A near real-time update window is set to retain the latest data and gradually eliminate old data.

[0085] Set a static feature baseline and calculate the mean and variance of each feature dimension within the window. If the feature dimension shows high stability, the variance is small and the mean fluctuates gently, indicating that the mobile device's Z-axis displacement on the ground is close to zero and the variance is approaching zero. If the feature dimension shows low stability, the variance is large or the mean changes suddenly, indicating bumpy ground. Finally, analyze the linkage relationship between features on different axes. For example, synchronized changes in the X- and Y-axis displacements of a mobile device indicate a vehicle turn, while isolated fluctuations on the Z-axis are often due to road bumps.

[0086] Depending on the scenario type, the threshold is raised on bumpy roads and set based on historical variance, and the threshold size is dynamically adjusted in real time. If the deviation of the feature vector exceeds the threshold, it is judged as noise or dynamic interference and is eliminated.

[0087] Finally, by setting the feature hierarchical retention, the confidence weights are allocated, including high-confidence features, medium-confidence features and low-confidence features.

[0088] in:

[0089] High-confidence features have low deviation and high historical consistency and are directly used for scene recognition.

[0090] Medium confidence features have medium deviation but are scene-relevant, and are weighted down for calculation.

[0091] Low confidence features: those with high deviation or isolated appearance are directly discarded.

[0092] Specifically, if the mobile device is traveling on a forest road, the stability characteristics are that the Z-axis displacement of the ground point is close to zero and the X / Y-axis angle changes smoothly.

[0093] Transient interference is caused by sensor jitter, which leads to a sudden increase in Z-axis displacement. If the deviation exceeds the threshold, it is directly filtered.

[0094] The statistical model response retains the stable characteristics of the ground, downweights the branch area, filters the instantaneous jitter data, and if high-frequency vibration is continuously detected, switches to the strong vibration template of the forest road and enables large receptive field filtering.

[0095] The spatial transformation matrix is ​​transformed into three-dimensional convolution kernel deformation parameters. Differentiated kernel deformation calculation is performed on the X, Y, and Z axes of the three-dimensional spatial rigid transformation field through a dynamic convolution kernel. The differentiated kernel deformation calculation includes vibration-sensitive axis feature processing and stable axis feature processing. Large receptive field smoothing filtering is used for the vibration-sensitive axis, and centralized high-resolution feature extraction is performed on the stable axis.

[0096] Specifically, a specific implementation method for establishing a spatial transformation matrix includes:

[0097] The rigid motion parameters after feature encoding are converted into deformation parameters of the three-dimensional convolution kernel, where the deformation parameters include kernel size, sampling interval, etc. According to the vibration characteristics of the scene, such as the Z-axis vibration caused by vehicle bumps, different deformation strategies are defined for the X, Y, and Z axes respectively.

[0098] Furthermore, different deformation strategies are defined for the X, Y, and Z axes, including:

[0099] If the vehicle bumps cause Z-axis vibration, a large-scale smoothing filter is used to expand the receptive field of the convolution kernel and suppress high-frequency jitter noise.

[0100] If the road edge or environmental conditions cause X and Y axis shaking, a high-resolution small convolution kernel is used to extract detailed features such as road edges and obstacle outlines.

[0101] The change amplitude of each axial deformation parameter is monitored in real time. If the vibration intensity exceeds the threshold, the dynamic convolution kernel is immediately frozen and switched to the pre-stored benchmark feature template. The scene recognition is quickly restored through template matching to avoid continuous error accumulation.

[0102] If the differential kernel deformation anomaly index exceeds the standard threshold, the dynamic convolution kernel is frozen, the baseline feature calculation is switched, and the pre-stored scene template is called for fast calibration.

[0103] The current scene information includes forest roads and mountain roads;

[0104] By receiving the current scene information, dynamically analyzing the current scene information, and outputting the analysis results to the strategy switching unit, the strategy switching unit dynamically outputs the forest road obstacle avoidance strategy and the mountain road obstacle avoidance strategy.

[0105] Furthermore, by analyzing environmental information such as the spatial distribution of vegetation to assist path planning, environmental factors are fully incorporated into path planning considerations, making the path planning of mobile devices more intelligent and reasonable, and able to better adapt to complex and changing actual environments.

[0106] Use laser transmitters to obtain 3D point cloud data of the mobile device environment and analyze the spatial distribution of vegetation;

[0107] Use radius filtering to remove noise points and separate ground points from vegetation points;

[0108] Specifically, radius filtering is a distance-based filtering method. For each point cloud data point, a data point is set as the center and a radius threshold is set. The number of points within this radius is counted. If the number of points is less than a preset threshold, it means that the area around the point is relatively sparse and is likely to be a noise point, which is removed. For example, for a point P, in a spherical area with a radius of r and centered on P, if the number of points is less than k, point P is considered to be a noise point and is removed.

[0109] Furthermore, the height method is used to separate ground points and vegetation points. The height of ground points is relatively low and flat, while the height of vegetation points is relatively high and has certain undulations. A height threshold h is set, and points with a height lower than h are regarded as ground points, and points with a height higher than h are regarded as vegetation points.

[0110] The 3D point cloud data is flattened into a 2D plane and divided into multiple grids to obtain the vegetation density corresponding to the number of point clouds in each grid. The vegetation density levels are divided according to the vegetation density, namely low density, passable, medium density, cautious passable, and high density, no passable.

[0111] Specifically, the 3D point cloud data is projected onto a 2D plane, the horizontal plane is selected, the height information of the point cloud data is ignored, and only the plane coordinates of the points are retained.

[0112] Furthermore, the flattened 2D plane is divided into multiple grids of the same size. Each grid is regarded as a small area for counting the number of point clouds. For example, the plane can be divided into square grids with a side length of s.

[0113] Furthermore, for each grid, the number of vegetation points falling within the grid is counted, and the vegetation density of each grid is calculated based on the area of ​​the grid and the number of vegetation points. The vegetation density is defined as the number of vegetation points per unit area. The calculated vegetation density is classified into three levels: low density (allowed to pass), medium density (allowed to pass with caution), and high density (allowed to pass). For example, different density thresholds are set for classification, such as the low density threshold is d1, and the medium density threshold is d2 (d2>d1). When the vegetation density is less than d1, it is low density (allowed to pass); when it is greater than or equal to d1 and less than d2, it is medium density (allowed to pass with caution); and when it is greater than or equal to d2, it is high density (allowed to pass).

[0114] Further, analyzing individual obstacles in the vegetation density, and identifying the height, width, and volume of the individual obstacles based on the acquired 3D point cloud data of the mobile device environment;

[0115] If the width and volume are greater than the maximum width that a mobile device can pass through, the mobile device will select a forced avoidance strategy. Low density can be prohibited by the strategy. If the width and volume are not greater than the maximum width that a mobile device can pass through, the mobile device will select a free avoidance strategy.

[0116] Vegetation density is divided into three levels: low-density passable, medium-density cautious passable, and high-density prohibited passable. This allows mobile devices to make appropriate decisions based on different vegetation conditions. Traditional methods may simply consider vegetation as an obstacle, but this hierarchical decision-making approach is more flexible and intelligent. For example, in low-density vegetation areas, mobile devices can pass directly, improving traffic efficiency; in medium-density areas, they must pass with caution, ensuring safety; and in high-density areas, they are prohibited from passing to avoid difficulties. In complex natural environments such as mountainous areas and forests, vegetation distribution and obstacle conditions are complex and diverse. This design comprehensively analyzes environmental information, including vegetation density and obstacle characteristics, enabling mobile devices to better adapt to various complex environments.

[0117] Through precise environmental perception and intelligent decision-making and planning, collisions between mobile devices and obstacles are effectively avoided, improving device safety. When encountering large obstacles, a forced avoidance strategy ensures that the device changes its path promptly to avoid danger. In areas with vegetation, decisions based on vegetation density can also reduce malfunctions or damage caused by entrapment.

[0118] The mountain road recognition analysis uses laser radar to obtain three-dimensional point cloud data of the terrain surface, removes abnormal points through radius filtering, and separates ground points from non-ground points through the RANSAC algorithm;

[0119] After data preprocessing is completed, the slope and curvature characteristics of the terrain are extracted, the steepness of the local terrain is analyzed, and the slope of each point cloud data point and its neighboring points is obtained;

[0120] The method for obtaining the slope of each point cloud data point and its adjacent points includes:

[0121] Calculate the elevation difference and horizontal distance between the first point cloud data point and each neighboring point.

[0122] The slope of each neighboring point of the first point cloud data point is obtained by calculating the inverse tangent trigonometric function of the elevation difference and horizontal distance between the first point cloud data point and each neighboring point.

[0123] Taking the average of the slopes of all neighboring points of the first point cloud data point as the slope value of the first point cloud data point;

[0124] The curvature of the first point cloud data point is calculated by the quotient of the second derivative of the slope and the slope value of the first point cloud data point.

[0125] Analyzing the steepness of the local terrain of the mobile device by obtaining the slope and curvature;

[0126] Analyzing the local terrain of the mobile device includes:

[0127] The terrain is divided into multiple local areas, which are segmented by setting multiple windows. The average slope and curvature are calculated in each window. The point cloud data of the current position of the mobile device is used to calculate its slope relative to that of neighboring points, determine the steepness of the current position, calculate the slope of each point on the path ahead, and predict the steepness of the future terrain. Combined with the slope information, the slope of the mobile device's path is minimized and steep areas are avoided. If the slope of the path ahead is detected to exceed the safety threshold, the system generates a warning message and replans the path.

[0128] Specifically, a certain number of points are randomly selected from the point cloud data. These points form a plane equation: ax+by+cz+d=0. The distances from the remaining points to the plane are calculated. A distance threshold is set. Points with distances less than the distance threshold are regarded as points on the plane, and the number of these points is counted. The above process is repeated N times, and each time a plane and the corresponding number of plane points are obtained. The plane with the largest number of plane points is selected, and the points belonging to the plane are marked as ground points, and the remaining points are non-ground points.

[0129] Furthermore, the entire terrain can be divided into several local areas. A regular grid division method can be used. For example, the terrain can be divided into multiple small areas with equal spacing in the x and y directions, or a backup storage structure can be set up to adaptively divide the local areas according to the complexity of the terrain, with finer divisions in areas with large terrain changes and coarser divisions in areas with relatively flat terrain.

[0130] Multiple rectangular windows are set in each local area. The window size is determined according to the actual situation. For example, the side length of the window can be set to a certain distance value, such as 1 meter, 2 meters, etc. In each rectangular window, the average slope and curvature of the point cloud data are calculated. The average slope is obtained by calculating the average slope value of all points in the window, and the average curvature is obtained in the same way.

[0131] Furthermore, based on the point cloud data point at the current location of the mobile device, the slope thereof and the slope of the adjacent points are calculated according to the above-mentioned method for calculating the slope. The calculated slope value is compared with a pre-set steepness level threshold to determine the steepness level of the current location, such as gentle, relatively steep, very steep, etc., and the slope of each point on the path ahead is calculated. The points on the path are selected according to the driving direction of the mobile device and a certain distance interval. Based on the slope values ​​of the points on the path, the future trend of terrain changes is analyzed and the steepness of the future terrain is predicted. For example, the average slope within a certain distance ahead is calculated, or the rate of change of the slope is analyzed to determine whether the terrain is becoming steeper or gentler.

[0132] The slope and curvature of each point cloud data point are accurately calculated, and based on this, the steepness of the local terrain is analyzed. Traditional methods may only consider the general undulations of the terrain and cannot provide such detailed terrain information. By calculating slope and curvature, mobile devices can better understand subtle changes in the surrounding terrain. For example, they can identify small slopes, bumps, or depressions, providing a richer basis for obstacle avoidance decisions. Local areas are adaptively divided according to the complexity of the terrain, with finer divisions in areas with large terrain changes and coarser divisions in areas with relatively flat terrain. The flexible division strategy enables the system to more efficiently process data from different terrains, improving the system's adaptability to various complex environments.

[0133] The mountain road identification analysis also includes identifying rocks and predicting rock falls;

[0134] Scan rock data using a laser transmitter and obtain rock geometric features, including height, width, and volume, through rock clustering. Rocks and rockfalls are then detected using a rock target recognition unit. The rock and rockfall recognition results are combined with the 3D point cloud data to analyze the mountain road and determine the location, size, and motion state of the rocks and rockfalls.

[0135] The location, size and movement status of the rocks and falling rocks are used to plan the path of the mobile equipment and generate early warning information.

[0136] A specific implementation method of rock clustering includes:

[0137] For each clustered rock, find the maximum value zmax and minimum value zmin of the Z-axis coordinate in the point cloud, then the height of the rock h=zmax-zmin.

[0138] Furthermore, to calculate the width of the circumscribed rectangle of the clustered rocks on a horizontal plane, such as the xy plane, we can first find the maximum value xmax and minimum value xmin of the x coordinate, as well as the maximum value ymax and minimum value ymin of the y coordinate, and then choose an appropriate method to calculate the width W according to the specific situation. Preferably, the calculation formula is: .

[0139] Furthermore, the volume is calculated by gridding, dividing the space where the rock point cloud is located into small grid cells, counting the number of points in each grid cell, and accumulating the points in the grid cells according to the cube of the side length of the grid cell to obtain the approximate volume of the rock.

[0140] A specific implementation method of the rock target recognition unit for identifying rocks or falling rocks includes:

[0141] Build an object recognition model based on a convolutional neural network in deep learning. Use a large number of labeled point cloud data samples of rocks and rockfalls for training. The samples should include rocks and rockfalls of different shapes, sizes, positions, and postures, and select an appropriate loss function for training.

[0142] Furthermore, based on the coordinate information of the point cloud of rocks and fallen rocks, the specific locations are determined in the three-dimensional space of the mountain road, and the location information is combined with the map data to intuitively display the locations of rocks and fallen rocks on the mountain road.

[0143] Furthermore, the geometric features obtained by calculation, including height, width, and volume, are used to further determine the size of rocks and rockfalls, and the size information is classified into small, medium, and large rocks and rockfalls.

[0144] Furthermore, multiple scans were performed at different time points, and the motion status of rocks and falling rocks was analyzed by comparing the point cloud data at different time points. For example, parameters such as the displacement, velocity, and acceleration of the point cloud were calculated. Through a feature matching method, corresponding feature points were found in the point cloud at different time points, and then the displacement of these feature points was calculated to obtain the motion information of rocks and falling rocks.

[0145] By performing multiple scans at different time points and calculating point cloud parameters such as displacement, velocity, and acceleration to analyze the motion of rocks and falling rocks, the system can dynamically understand their movement trends in real time and predict their possible trajectories in advance, enabling mobile devices to proactively avoid obstacles and improving the timeliness and effectiveness of the obstacle avoidance system. Using a feature-matching method to find corresponding feature points in the point cloud at different time points and calculate displacement, the system can more accurately track the movement of rocks and falling rocks compared to traditional simple comparisons or assumptions. This precise acquisition of motion information helps mobile devices more flexibly adjust their driving paths to avoid moving rocks and falling rocks, enhancing the obstacle avoidance system's ability to cope with dynamic obstacles.

[0146] The strategy switching unit intelligently switches to the obstacle avoidance strategy for the current scene based on the recognition results;

[0147] A specific implementation method of a strategy switching unit includes:

[0148] The inputs of the strategy switching unit include:

[0149] Forest road analysis results, including vegetation density; geometric characteristics of individual obstacles; mountain analysis results, including terrain steepness; location, size and movement of rocks and falling rocks.

[0150] The output of the strategy switching unit includes different obstacle avoidance strategies, including:

[0151] Mandatory avoidance strategy: the device must avoid obstacles or dangerous areas.

[0152] Free avoidance strategy, the device can selectively avoid obstacles.

[0153] Path optimization strategy: the device adjusts its path to avoid dangerous areas.

[0154] The generation logic of different obstacle avoidance strategies includes:

[0155] The density level is determined through vegetation density analysis, and the corresponding obstacle avoidance strategy is determined based on the density level. Specifically, in low-density areas, a free avoidance strategy is generated, allowing the device to pass freely and selectively avoid obstacles.

[0156] In medium-density areas, a path optimization strategy is generated, and the device needs to slow down and adjust its path.

[0157] A mandatory avoidance strategy is generated for high-density areas, and the device must avoid the area.

[0158] Individual obstacle analysis includes: if the width and volume of the obstacle are greater than the maximum width of the device, a mandatory avoidance strategy is generated; if the width and volume of the obstacle are less than the maximum width of the device, a free avoidance strategy is generated.

[0159] The avoidance strategy for mountain roads is determined by the steepness level, including: generating a free avoidance strategy in flat areas, and the equipment can pass freely.

[0160] Generate a route optimization strategy in medium-sized areas, and the device needs to slow down and adjust the route.

[0161] Generates a mandatory avoidance maneuver in steep areas, and the device must avoid the area.

[0162] Rock and rockfall analysis generates a mandatory avoidance strategy when rocks or rockfalls are detected. The equipment must avoid the danger zone. Based on the movement state of rocks and rockfalls, its future position is predicted and a path optimization strategy is generated.

[0163] A strategy switching unit workflow includes:

[0164] Receive analysis results for forest roads and mountain roads;

[0165] Generate forest road avoidance strategies based on vegetation density and individual obstacle analysis results;

[0166] Generate mountain road avoidance strategies based on terrain steepness and rock and rockfall analysis results;

[0167] Combined with avoidance strategies, equipment paths are optimized to ensure safety and feasibility.

[0168] Generate early warning information for dangerous situations such as high-density vegetation, steep terrain, rocks and falling rocks.

[0169] Output the generated avoidance strategy and warning information to the path planning module and the actuator;

[0170] The movement parameters of the mobile device are controlled through the path planning module.

[0171] Example 2:

[0172] A laser obstacle avoidance system for a mobile device, wherein the specific implementation method further includes:

[0173] If a mobile device is traveling on a forest road in a forest farm, a specific implementation method for obstacle avoidance includes:

[0174] The maximum bending force of plant branches is obtained through the vegetation density and vegetation type. If the interaction force generated by the mobile device passing through the forest road and contacting the branches exceeds the bearing limit of the branches, the branches will break; if the force is too large, the mobile device will be avoided.

[0175] A specific implementation method for plant branch bending and mobile device obstacle avoidance includes: generating point cloud density through solid-state LiDAR.

[0176] Equipped with visual sensors to generate depth maps.

[0177] Use radar sensors to detect tiny vibrations in vegetation branches and determine their rigidity characteristics.

[0178] Establish a database of mechanical parameters of common forest plants, including elastic modulus, bending strength, etc.

[0179] The branches were simplified into cantilever beams, and the critical fracture force of the branches was obtained from the mechanical parameter database of forest plants.

[0180] When the contact force is detected to be critical to the fracture force, an avoidance path is generated while limiting the acceleration of the device to reduce the impact force.

[0181] like Figure 4 As shown, the branches of the plants block the movement of the mobile device. The mobile device applies force to the branches of the plants, and the branches of the plants react to the mobile device. If the plants are lush, the reaction force of the branches of the plants is large, which hinders the normal movement of the mobile device and even breaks and damages the mobile device. Therefore, by detecting the vibration of the branches and the density of the plants, the mobile device can form an obstacle avoidance strategy.

[0182] The mobile device includes a small-diameter gear and a large-diameter gear. The roll angle of the mobile device is obtained by predicting the slope of the point. If the roll angle of the mobile device does not exceed the maximum load threshold of the device, the mobile device switches to the small-diameter gear output power and adjusts the motor output torque to control the rotation of the small-diameter gear. The output power of the small-diameter gear drives the large-diameter gear to rotate through the transmission device, thereby controlling the speed of the mobile device over the slope.

[0183] A specific implementation method for slope obstacle avoidance of a mobile device includes:

[0184] Obtain device location via RTK-GPS and match it against an open source database.

[0185] The local elevation map is generated using LiDAR point cloud, which is aligned with the open source database through an iterative nearest neighbor algorithm to predict the forward slope.

[0186] Furthermore, a two-speed transmission is designed for mobile equipment, specifically including:

[0187] Small diameter gears include module and number of teeth and are used for high torque output.

[0188] Large diameter gears include module and number of teeth and are used for high-speed flat road driving.

[0189] The brushless motor is controlled through field-oriented control, and the MCU adjusts the dq axis current in real time to improve the torque control accuracy.

[0190] Adjust the roll angle through the center of gravity, activate the hydraulic suspension system, and adjust the height difference between the left and right wheels.

[0191] Through the coordinated control of gear switching and center of gravity adjustment, the maximum climbing angle and roll stability of the equipment are improved.

[0192] like Figure 2 As shown, as the slope change of the mobile device increases, the small-diameter gear linkage transmission device is switched to drive the large-diameter gear, which is used to reduce the moving speed and accurately control the mobile device to pass through the area with a larger slope.

[0193] It is important to note that the configuration and arrangement of the present application, as illustrated in various exemplary embodiments, are illustrative only. Although only two embodiments are described in detail in this disclosure, those reading this disclosure will readily appreciate that numerous modifications are possible without materially departing from the novel teachings and advantages of the subject matter described herein. For example, variations in the size, dimensions, structure, shape, and proportions of various components, as well as parameter values ​​(e.g., temperature, pressure, etc.), mounting arrangements, use of materials, color, orientation, and the like, are possible. For example, components shown as integrally formed may be comprised of multiple parts or components, the positions of components may be inverted or otherwise altered, and the nature, number, or position of discrete components may be modified or changed. Therefore, all such modifications are intended to be encompassed within the scope of this invention. The order or sequence of any process or method steps may be altered or reordered according to alternative embodiments. Any "means-plus-function" clause is intended to cover structures that perform the functions described herein, and not only structural equivalence but also structural equivalents. Other substitutions, modifications, changes, and omissions may be made in the design, operating conditions, and arrangement of the exemplary embodiments without departing from the scope of this invention.

[0194] Additionally, in order to provide a concise description of exemplary embodiments, all features of an actual embodiment may not be described (i.e., those features that are not relevant to the best mode presently contemplated for carrying out the invention or those that are not relevant to implementing the invention).

[0195] It should be understood that in the development of any actual embodiment, as in any engineering or design project, numerous implementation-specific decisions may be made. Such a development effort may be complex and time-consuming, but for those of ordinary skill having the benefit of this disclosure, the development effort will be a routine task of design, fabrication, and production without undue experimentation.

[0196] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, and all of these should be included in the scope of the present invention.

Claims

1. A laser obstacle avoidance system for a mobile device, characterized in that: include: Multi-line laser scanning emits multiple parallel laser beams and receives reflected light to obtain three-dimensional information, processes the three-dimensional information, and generates a preliminary obstacle avoidance strategy; The multi-line laser scanning comprises: The laser transmitter is used to emit multiple parallel laser beams spaced at a preset distance; The laser receiver is used to receive the laser beam emitted by the laser transmitter and reflected by the surrounding environment, and convert the received laser beam into an electrical signal; The scanning mechanism is used to drive the laser emitter and the laser receiver to rotate or translate within a preset range; The data processing module is used to process the electrical signal converted by the laser receiver to generate data representing the three-dimensional structure of the surrounding environment; The scene recognition module receives the processed three-dimensional information in real time, identifies the current application scene, and outputs a classification result of the current scene information; Inputting the processed three-dimensional coordinates of the three-dimensional information into a three-dimensional space rigid transformation field, wherein the three-dimensional space rigid transformation field constructs a rigid motion pattern of a neighborhood of an object on a forest road or a mountain road; Extracting eigenvalues ​​of the rigid motion mode of the neighborhood, encoding the eigenvalues ​​into a spatial transformation matrix, evaluating the accuracy of the spatial transformation matrix, sorting by the accuracy, and retaining stable features of the current application scene acquired by the laser sensor; The scene recognition module is used to classify different environmental scenes and output current scene information; The strategy switching unit receives the classification result of the current scene information and dynamically outputs the obstacle avoidance strategy of the current scene; The current scene information includes forest roads and mountain roads; By receiving the current scene information, dynamically analyzing the current scene information, outputting the analysis result to the strategy switching unit, and the strategy switching unit dynamically outputting the forest road obstacle avoidance strategy and the mountain road obstacle avoidance strategy; The actuator controls the movement of the mobile device according to the obstacle avoidance strategy.

2. The laser obstacle avoidance system for mobile devices according to claim 1, characterized in that: The spatial transformation matrix is ​​transformed into three-dimensional convolution kernel deformation parameters. Differentiated kernel deformation calculation is performed on the X, Y, and Z axes of the three-dimensional spatial rigid transformation field through a dynamic convolution kernel. The differentiated kernel deformation calculation includes vibration-sensitive axis feature processing and stable axis feature processing. Smoothing filtering is applied to the vibration-sensitive axis, and feature extraction is performed on the stable axis. If the differential kernel deformation anomaly index exceeds the standard threshold, the dynamic convolution kernel is frozen, the baseline feature calculation is switched, and the pre-stored scene template is called for calibration.

3. The laser obstacle avoidance system for mobile devices according to claim 2, characterized in that: Dynamic analysis of forest paths through vegetation density analysis and individual plant growth; Use laser transmitters to obtain 3D point cloud data of the mobile device environment and analyze the spatial distribution of vegetation; Use radius filtering to remove noise points and separate ground points from vegetation points; Flatten the 3D point cloud data and divide it into multiple grids, obtain the vegetation density corresponding to the number of point clouds in each grid, and classify the vegetation density levels according to the vegetation density, namely low density, passable, medium density, cautious passable, and high density, no passable; Individual obstacles in the vegetation density are analyzed, and the height, width, and volume of the individual obstacles are identified based on the acquired 3D point cloud data of the mobile device environment.

4. The laser obstacle avoidance system for mobile devices according to claim 3, characterized in that: The mountain road recognition analysis uses laser radar to obtain three-dimensional point cloud data of the terrain surface, removes abnormal points through radius filtering, and separates ground points from non-ground points; Extract the slope and curvature characteristics of the terrain, analyze the steepness of the local terrain, and obtain the slope of each point cloud data point and its neighboring points; The method for obtaining the slope of each point cloud data point and its adjacent points includes: Calculate the elevation difference and horizontal distance between the first point cloud data point and each adjacent point; Obtaining the slope of each neighboring point of the first point cloud data point by calculating the inverse tangent trigonometric function of the elevation difference and horizontal distance between the first point cloud data point and each neighboring point; Taking the average of the slopes of all neighboring points of the first point cloud data point as the slope value of the first point cloud data point; The curvature of the first point cloud data point is calculated by the quotient of the second derivative of the slope and the slope value of the first point cloud data point.

5. The laser obstacle avoidance system for mobile devices according to claim 4, characterized in that: Analyzing the steepness of the local terrain of the mobile device by obtaining the slope and curvature; Analyzing the local terrain of the mobile device includes: Divide the terrain into multiple local areas, segment the multiple local areas by setting multiple windows, and calculate the average slope and curvature in each window; Starting from the current position of the mobile device, a prediction point is set at a preset distance along the direction of movement of the mobile device. The slope of each prediction point is calculated based on the current slope and curvature. If the curvature is positive, the slope gradually increases, and if the curvature is negative, the slope gradually decreases. If the terrain is bumpy, the window size is increased; if the terrain is flat, the window size is reduced. There are multiple collection points in the local area. If the slope of a collection point exceeds the set threshold with the slope of the previous prediction point, local resampling is started to recalculate the slope of the current mobile device location.

6. The laser obstacle avoidance system for mobile devices according to claim 5, characterized in that: The mountain road identification analysis also includes identifying rocks and predicting rock falls; Rock data is scanned by a laser transmitter, and the geometric characteristics of the rocks, including height, width and volume, are obtained through rock clustering. Rocks and fallen rocks are detected by a rock target recognition unit. The recognition results of rocks and fallen rocks are combined with the 3D point cloud data to analyze the mountain road and determine the position, size and movement state of the rocks and fallen rocks.

7. The laser obstacle avoidance system for mobile devices according to claim 6, characterized in that: The mobile device includes a small-diameter gear and a large-diameter gear. The roll angle of the mobile device is obtained by predicting the slope of the point. If the roll angle of the mobile device does not exceed the maximum load threshold of the device, the mobile device switches to the small-diameter gear output power and adjusts the motor output torque to control the rotation of the small-diameter gear. The output power of the small-diameter gear drives the large-diameter gear to rotate through the transmission device, thereby controlling the speed of the mobile device over the slope.

Images