Obstacle sensing method and system based on laser radar data analysis

Abstract

The invention discloses an obstacle sensing method and system based on laser radar data analysis, and relates to the technical field of radar data obstacle analysis, and the method comprises the steps: scanning a surrounding environment through a multi-mode sensor, obtaining sensor data, synchronously collecting the data of the multi-mode sensor, and carrying out the preprocessing, an obstacle target is obtained from a camera image, sensor data are integrated by using a data fusion algorithm, and obstacle features are extracted by using a segmentation threshold function based on density clustering in combination with an obstacle local feature descriptor and global feature description to form complete obstacle feature description. The laser radar can provide high-resolution distance data, so that the system can accurately identify and position obstacles in the surrounding environment, algorithm processing is combined, the method can complete scanning and analysis of the environment in a short time, rapid response to the obstacles in the dynamic environment is achieved, and the system can be widely applied to the field of dynamic environment monitoring. And the accuracy and reliability of obstacle detection are greatly improved.

Description

[0001] The present invention relates to the technical field of radar data analysis for obstacles, and particularly to an obstacle perception method and system based on lidar data analysis. Background Art

[0002] With the rapid development of autonomous driving technology and intelligent robot applications, the environmental perception system, as one of its core technologies, lidar, as a high-precision distance measurement device, has gradually become a key sensor in environmental perception due to its strong anti-interference ability and high-resolution characteristics in complex environments. By generating three-dimensional cloud data of the surrounding environment, the problems of lagging obstacle avoidance reaction and over-conservatism not being effectively solved due to the deviation of the data collected by lidar sensors in time and during fusion are addressed.

[0003] Although existing lidar data analysis methods have made significant progress in the field of environmental perception, there are still certain limitations in the performance of existing technologies in predicting the movement trajectories of obstacles. The traditional Kalman filter can estimate the state changes of linear systems well, but has poor prediction effects for non-linear systems. This may lead to inaccurate prediction of the future positions of obstacles in practical applications, thereby affecting the effectiveness of obstacle avoidance strategies. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an obstacle perception method based on lidar data analysis to solve the problems of lagging obstacle avoidance reaction and over-conservatism not being based on obstacle types due to the deviation of the data collected by lidar sensors in time and during fusion.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides an obstacle perception method based on lidar data analysis, which includes: Scanning the surrounding environment using a multi-modal sensor to obtain sensor data, and synchronously collecting and preprocessing the data of the multi-modal sensor; Obtaining obstacle targets from camera images and integrating sensor data using a data fusion algorithm; Using a segmentation threshold function based on density clustering in combination with obstacle local feature descriptors and global features to extract obstacle features and form a complete obstacle feature description; Using a deep learning model CNN to construct a classification model, detecting obstacle features from the fused sensor data, and classifying the detected obstacles using the classification model; Using a Kalman filter to predict the movement trajectory of obstacles in advance and formulating corresponding obstacle avoidance strategies according to the types of obstacles.

[0007] As the obstacle perception method based on lidar data analysis according to the present invention, wherein: Use a multi-modal sensor to scan the surrounding environment to obtain sensor data, and synchronously collect and preprocess the data of the multi-modal sensor. Specifically, The synchronous collection means that the lidar emits laser pulses and receives reflected signals to generate high-density three-dimensional cloud data. The millimeter-wave radar detects the distance and speed of objects to supplement the deficiencies of the lidar under rainy and snowy conditions. The camera captures environmental images to provide texture information; The preprocessing means that the bilateral filter is used to remove noise from the collected lidar three-dimensional cloud data.

[0008] As the obstacle perception method based on lidar data analysis according to the present invention, wherein: Obtain obstacle targets from camera images and use a data fusion algorithm to integrate sensor data. Specifically, Use the deep learning model YOLOv5 to preprocess the camera images and locate the obstacles in the images; Align the three-dimensional cloud data generated by the lidar with the camera images. The millimeter-wave radar provides the radial velocity of the obstacles. Combine the lidar points and the obstacle target positions in the camera images to obtain the velocity vector of each target; Use the Kalman filter to fuse the multi-modal sensor data into one.

[0009] As the obstacle perception method based on lidar data analysis according to the present invention, wherein: Use a segmentation threshold function based on density clustering to combine the local feature descriptors and global feature descriptors of the obstacles to extract the obstacle features and form a complete obstacle feature description. Specifically, Use the segmentation threshold function algorithm to perform clustering segmentation on the three-dimensional cloud data. For each clustered obstacle segment, use the FPFH feature histogram to extract local features, and use the PointNet model to extract the global features of each obstacle segment; Combine the local features and global features according to the simple average calculation method to obtain the overall shape and spatial distribution of the obstacles.

[0010] As the obstacle perception method based on lidar data analysis according to the present invention, wherein: Use the deep learning model CNN to construct a classification model. Specifically, Select the PointNet model as the convolutional neural network CNN architecture for three-dimensional cloud data classification and optimize it using the Adam optimizer; Forward-propagate the multi-modal sensor data through the convolutional neural network of the PointNet model to calculate the predicted values, calculate the loss function value based on the predicted values and the true labels, calculate the gradients through the backpropagation algorithm and update the model parameters, and repeat the above steps until the model reaches the predetermined number of training rounds.

[0011] As the obstacle perception method based on lidar data analysis according to the present invention, wherein: Perform obstacle feature detection on the fused sensor data, and use a classification model to classify the detected obstacles. Specifically, Preprocess the sensor data that has been integrated by the Kalman filter, use the DBSCAN algorithm to perform clustering segmentation on the three-dimensional cloud data, identify different obstacle segments, use the FPFH fast point feature histogram to extract the obstacle features for each clustered obstacle segment, input the extracted obstacle features into the trained classification model, the model outputs the probability distribution of each obstacle, and select the category with the highest probability as the classification result of the obstacle.

[0012] As the obstacle perception method based on lidar data analysis according to the present invention, wherein: Use the Kalman filter to predict the motion trajectory of the obstacle in advance, and formulate corresponding obstacle avoidance strategies according to the obstacle type. Specifically, Use the Kalman filter transition matrix to describe the evolution of the obstacle state over time, predict multiple future time steps through the Kalman filter covariance matrix, generate the future motion trajectory of the obstacle, and select the corresponding obstacle avoidance strategy according to the results of the previous classification model.

[0013] In a second aspect, the present invention provides an obstacle perception system based on lidar data analysis, including, A data preprocessing module, a data fusion module, a feature extraction module, a classification model module, and a filter module. Specifically, The preprocessing module identifies and removes random noise by dynamic noise filtering, unifies the data reference framework through coordinate system conversion, and reduces the data volume by point cloud compression technology to improve efficiency; The data fusion module solves the sensor movement difference by multi-temporal frame spatial alignment, and enhances the information expression by fusing the feature vectors of different data sources at the feature level; The feature extraction module calculates the FPFH descriptor to capture the geometric characteristics of the point cloud, analyzes the reflection intensity to provide clues about the surface properties of the object, and assists in distinguishing different objects; The classification model module divides the point cloud by PointNet++ to learn local to global features, and adaptively adjusts the weights by dynamic convolution to improve the adaptability to diverse three-dimensional cloud data; The filter module initializes the state vector of the improved Kalman filter and accurately estimates the system motion state through the prediction and update loop.

[0014] In a third aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and: when the computer program is executed by the processor, any step of the obstacle perception method based on lidar data analysis as described in the first aspect of the present invention is implemented.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and: when the computer program is executed by the processor, any step of the obstacle perception method based on lidar data analysis as described in the first aspect of the present invention is implemented.

[0016] The beneficial effects of the present invention are as follows: The lidar can provide high-resolution distance data, enabling the system to accurately identify and locate obstacles in the surrounding environment. Since the data acquisition speed of the lidar is very fast and combined with efficient algorithm processing, this method can complete the scanning and analysis of the environment in a short time, achieving a fast response to obstacles in a dynamic environment, which greatly improves the accuracy and reliability of obstacle detection. Different from traditional vision sensors that rely on lighting conditions, the lidar can work stably under various weather conditions, enhancing robustness and adaptability. This method can not only identify the presence of obstacles but also infer the approximate type of obstacles based on information such as distance and shape, providing an important basis for subsequent behavior planning and decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts.

[0018] Figure 1 It is a flowchart of the obstacle perception system architecture of the multi-modal sensor in Embodiment 1.

[0019] Figure 2 It is a schematic diagram of the synchronous acquisition of multi-modal sensor data in Embodiment 1.

[0020] Figure 3 It is a schematic diagram of feature extraction and classification processing in Embodiment 1.

[0021] Figure 4 It is a schematic diagram of motion trajectory prediction and obstacle avoidance strategy generation in Embodiment 1. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] In order to make the above objects, features, and advantages of the present invention more obvious and understandable, the following will provide a detailed description of the specific embodiments of the present invention in conjunction with the accompanying drawings of the specification.

[0023] In the following description, many specific details are set forth in order to fully understand the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0024] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation manner of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that mutually excludes other embodiments.

[0025] Embodiment 1, referring to Figures 1 to 4 , is the first embodiment of the present invention. This embodiment provides an obstacle perception method based on lidar data analysis, including the following steps: S1. Use a multi-modal sensor to scan the surrounding environment to obtain sensor data, and synchronously collect and preprocess the data of the multi-modal sensor, including the following steps. The lidar device emits laser pulses at a transmission frequency. The lidar receiver receives the laser signals reflected from the object surface. According to the time difference between transmission and reception, the distance of each point is calculated, and combined with the attitude information of the lidar such as the angle, high-density three-dimensional cloud data is generated. The millimeter-wave radar emits millimeter-wave signals at a frequency. The millimeter-wave radar receiver receives the millimeter-wave signals reflected from the object surface. By analyzing the time difference between the transmitted and received signals and applying the Doppler effect, the distance and speed of the target object are calculated. The camera captures two-dimensional images of the environment at a frame rate of 30 fps. A clock synchronization mechanism is used to ensure the temporal synchronization of the data of the lidar, millimeter-wave radar, and camera, and transfer the data of different sensors to the same coordinate system to ensure the spatial consistency of all sensor data, facilitating subsequent fusion and processing.

[0026] In order to remove the noise in the lidar three-dimensional cloud data, a bilateral filter is used for smoothing. The bilateral filter not only considers the spatial distance but also the intensity difference, retains important edge information, and at the same time reduces the influence of noise. Specifically, the influence range of the domain diameter d on any point in the image is a square area with a side length of d. Usually, this square area is an odd number to have a clear center point. A larger diameter means a larger neighborhood, which can better smooth the noise.

[0027] S2, obtain obstacle targets from camera images and integrate sensor data using data fusion algorithms, including the following steps: The 2D image captured by the camera is input into the YOLOv5 model for forward propagation to generate a feature map, and non-maximum suppression is applied to the feature map to filter redundant detection boxes, and finally the position and category information of each detection box is output. According to the output of the YOLOv5 model, the bounding box coordinates of each obstacle target are obtained. The 3D cloud data generated by the lidar is transferred from the lidar coordinate system to the camera coordinate system, and the transferred 3D cloud data is projected onto the image plane of the camera. According to the position of the projection point, the camera image is matched with the bounding box detected by the YOLOv5 model. The IoU can be used as the matching criterion to find the closest bounding box.

[0028] The radial velocity of each detected target is extracted from the millimeter-wave radar. The velocity vector of each target is calculated using the target position information in the lidar 3D cloud data and the camera image, combined with the radial velocity provided by the millimeter-wave radar. The radial velocity provided by the millimeter-wave radar is combined with the target's motion direction in the image plane to obtain a complete velocity vector.

[0029] The data from the lidar, camera and millimeter-wave radar are used as observation values and input into the Kalman filter for fusion. By fusing the data of multiple sensors, more abundant sensor information can be obtained, which significantly improves the accuracy and reliability of obstacle detection. In particular, the advantage of millimeter-wave radar in adverse weather conditions further enhances the robustness of radar sensors, and can obtain high-precision position and speed information of each target, improving the accuracy of target tracking.

[0030] S3, using the segmentation threshold function based on density clustering combined with the obstacle local feature descriptor and the global feature description to extract obstacle features to form a complete obstacle feature description, including the following steps: DBSCAN was chosen as the clustering algorithm because it performs well in dealing with irregular shapes and noisy data. Based on the two key parameters of DBSCAN, neighborhood radius and minimum point, the points in the neighborhood are calculated and all points in the neighborhood are classified into the same cluster. If there are new points in the neighborhood, the cluster is expanded until no new points can be added. Points that do not belong to any cluster are marked as noise points.

[0031] After DBSCAN clustering, multiple obstacle fragments are obtained. The principal component analysis method is used to analyze the normal vector of each point. All neighboring points are found through each point and the neighborhood radius. The FPFH reinforcement learning algorithm is used to calculate the distance and angle between each point to obtain a set of local features. The PointNet model is selected as the global feature extraction model because the PointNet model can directly process three-dimensional cloud data and extract high-level global features. The PointNet model consists of a multi-layer perceptron in the input layer and a maximum pooling layer. The three-dimensional cloud data is input into the PointNet model to pre-train the PointNet model. The clustered fragments are input into the PointNet model for forward propagation to generate the global feature vector of each clustered fragment, and the local features and global features are concatenated to form a comprehensive feature vector. The local features and global features are weighted averaged using the simple averaging method, so that the final feature vector contains both local detail information and global structure information. For each clustered fragment, the overall shape and spatial distribution of the obstacle are obtained.

[0032] S4. Use the deep learning model CNN to build a classification model, including the following steps: Use the PyTorch framework to load the pre-trained PointNet model, which consists of a multi-layer perceptron in the input layer and a maximum pooling layer. Select the Adam optimizer as the optimizer in the training process because the Adam optimizer performs well when processing large-scale data sets and has the advantage of adaptive learning rate. Preprocess the 3D cloud data, camera images, and millimeter-wave radar data generated by the lidar and combine them together to form a comprehensive feature vector, which is input into the PointNet model for final classification prediction and outputs the predicted value. Use one-hot encoding as the true label, and use the predicted value and the true label to calculate the cross entropy loss. Use PyTorch's automatic differentiation function to calculate the gradient of the loss function with respect to the model parameters, and use the Adam optimizer to update the model parameters based on the calculated gradient. Repeat the above steps of forward propagation, loss calculation, back propagation, and parameter update until the model reaches the predetermined 6 training rounds.

[0033] S5, performing obstacle feature detection on the fused sensor data and classifying the detected obstacles using a classification model, including the following steps: Use the DBSCAN algorithm to cluster and segment 3D cloud data. For the predicted probability distribution of each clustering segment, select the category with the highest probability as the classification result of the current obstacle. Load the previously trained PointNet classification model, input the FPFH feature vector of each clustering segment into the PointNet model. An important feature of the PointNet model is that it can extract global feature vectors from local geometric structures. The last layer of the PointNet model is a fully connected layer combined with a softmax activation function for calculation, outputting the probability distribution of each category, outputting the probability distribution of each obstacle. For the predicted probability distribution of each clustering segment, select the category with the highest probability as the classification result of the obstacle, output the classification result of each clustering segment, and associate the classification result with the original 3D cloud data for subsequent processing.

[0034] S6. Use a Kalman filter to predict the motion trajectory of the obstacle in advance and formulate corresponding obstacle avoidance strategies according to the type of the obstacle, including the following steps. The state of the obstacle is described by three parameters: position, velocity, and acceleration. The state vector is used to represent the current position, velocity, and acceleration of the obstacle. By inputting the state vector into the linear dynamic model, an obstacle whose state evolves over time is obtained.

[0035] The linear dynamic model is based on a state transition matrix as its basic structure. The state transition matrix describes how the obstacle transfers from the current state to the next state. According to the Kalman filter, initialize the initial state matrix and the initial covariance matrix. Use the state transition equation for prediction at each time point. To predict multiple future time steps, repeat the above prediction steps multiple times. Each prediction will update the state estimate and the covariance matrix. Extract the state vector from each step of the prediction to generate a series of predicted values of future states. Extract the position component of the state vector from each step of the prediction. The position component represents the future motion trajectory of the obstacle. The position component can be used to draw the future path map of the obstacle to visually display the motion trend of the obstacle. Plot the predicted position data into a graph to more intuitively understand the future motion of the obstacle. Through the classification model, the category of the obstacle is obtained, such as the obstacle may be a pedestrian, a vehicle, or other static objects. According to the specific type of the obstacle, select the obstacle avoidance strategy. If the obstacle is a pedestrian, adopt a more conservative obstacle avoidance strategy, such as decelerating and maintaining a safe distance. If the obstacle is a static object, calculate whether to adjust the path and continue to move forward according to the speed and trajectory of the vehicle, and set decision rules to ensure that the optimal decision can be made in different situations.

[0036] This embodiment also provides an obstacle perception system based on lidar data analysis, including: The preprocessing module identifies and removes random noise through dynamic noise filtering, converts the coordinate system to unify the data reference framework, and uses point cloud compression technology to reduce the data volume and improve efficiency; The data fusion module aligns the multi-temporal frames in space to solve the sensor movement differences, and fuses the feature vectors of different data sources at the feature level to enhance the information expression; The feature extraction module calculates the FPFH descriptor to capture the geometric characteristics of the point cloud, analyzes the reflection intensity to provide clues about the surface properties of the object, and helps to distinguish different objects; The classification model module divides the point cloud by PointNet++ to learn local to global features, and adaptively adjusts the weights through dynamic convolution to improve the adaptability to diverse 3D cloud data; The filter module initializes the state vector through an improved Kalman filter and accurately estimates the system motion state through the prediction and update loop.

[0037] This embodiment also provides a computer device applicable to a situation of an obstacle perception method based on lidar data analysis, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement an obstacle perception method based on lidar data analysis as proposed in the above embodiment.

[0038] This computer device can be a terminal. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the shell of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0039] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the obstacle perception method based on lidar data analysis as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, abbreviated as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, abbreviated as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, abbreviated as EPROM), programmable read-only memory (Programmable Red-Only Memory, abbreviated as PROM), read-only memory (Read-Only Memory, abbreviated as ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0040] In summary, in the present invention: The lidar can provide high-resolution distance data, enabling the system to accurately identify and locate obstacles in the surrounding environment. Since the data acquisition speed of the lidar is very fast, combined with efficient algorithm processing, this method can complete the scanning and analysis of the environment in a short time, achieving a fast response to obstacles in a dynamic environment, which greatly improves the accuracy and reliability of obstacle detection. Different from traditional vision sensors that rely on lighting conditions, the lidar can work stably under various weather conditions, enhancing the robustness and adaptability. This method can not only identify the presence of obstacles, but also infer the approximate type of obstacles based on information such as distance and shape, providing an important basis for subsequent behavior planning and decision-making.

[0041] 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 them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. An obstacle perception method based on lidar data analysis, characterized in that: Including: Using a multi-modal sensor to scan the surrounding environment to obtain sensor data, and synchronously collecting the data of the multi-modal sensor for preprocessing; Obtaining obstacle targets from camera images and integrating sensor data using a data fusion algorithm; Using a segmentation threshold function based on density clustering, combined with local and global feature descriptors of obstacles, to extract obstacle features and form a complete obstacle feature description; Using a deep learning model CNN to construct a classification model, detecting obstacle features from the fused sensor data, and classifying the detected obstacles using the classification model; Using a Kalman filter to predict the motion trajectory of obstacles in advance and formulating corresponding obstacle avoidance strategies according to the types of obstacles.

2. The obstacle perception method based on lidar data analysis according to claim 1, wherein: Using a multi-modal sensor to scan the surrounding environment to obtain sensor data, and synchronously collecting the data of the multi-modal sensor for preprocessing. Specifically, The synchronous collection means that the lidar emits laser pulses and receives reflected signals to generate high-density three-dimensional point cloud data, the millimeter-wave radar detects the distance and speed of objects to supplement the deficiencies of the lidar under rainy and snowy conditions, and the camera captures environmental images to provide texture information; The preprocessing means using a bilateral filter to remove noise from the collected lidar three-dimensional cloud data.

3. The obstacle perception method based on lidar data analysis according to claim 2, wherein: Obtaining obstacle targets from camera images and integrating sensor data using a data fusion algorithm. Specifically, Using the deep learning model YOLOv5 to preprocess the camera image and locate the obstacles in the image; Aligning the three-dimensional cloud data generated by the lidar with the camera image, the millimeter-wave radar provides the radial velocity of the obstacle, and combining the lidar points and the obstacle target positions in the camera image to obtain the velocity vector of each target; Using a Kalman filter to fuse the multi-modal sensor data into one.

4. The obstacle perception method based on lidar data analysis according to claim 3, wherein: Using a segmentation threshold function based on density clustering, combined with local and global feature descriptors of obstacles, to extract obstacle features and form a complete obstacle feature description. Specifically, Using the segmentation threshold function algorithm to cluster and segment the three-dimensional cloud data. For each clustered obstacle segment, using the FPFH feature histogram to extract local features and using the PointNet model to extract global features of each obstacle segment; Combining the local features and global features according to a simple average calculation method to obtain the overall shape and spatial distribution of the obstacle.

5. The obstacle perception method based on lidar data analysis according to claim 4, wherein: Using a deep learning model CNN to construct a classification model. Specifically, Selecting the PointNet model as the convolutional neural network CNN architecture for three-dimensional cloud data classification and optimizing it using the Adam optimizer; Feeding the multi-modal sensor data through the convolutional neural network of the PointNet model to calculate the predicted value, calculating the loss function value based on the predicted value and the true label, calculating the gradient through the backpropagation algorithm and updating the model parameters, and repeating the above steps until the model reaches the predetermined number of training epochs.

6. The obstacle perception method based on lidar data analysis according to claim 5, wherein: Detecting obstacle features from the fused sensor data and classifying the detected obstacles using the classification model. Specifically, Preprocess the sensor data that has been integrated by the Kalman filter, use the DBSCAN algorithm to cluster and segment the 3D cloud data, identify different obstacle segments, extract obstacle features for each clustered obstacle segment using the FPFH feature histogram, and input the extracted obstacle features into the trained classification model. The model outputs the probability distribution of each obstacle, and the category with the highest probability is selected as the classification result of the obstacle.

7. The obstacle perception method based on lidar data analysis according to claim 6, characterized in that: Use the Kalman filter to predict the motion trajectory of the obstacle in advance and formulate corresponding obstacle avoidance strategies according to the obstacle type. Specifically, Use the Kalman filter transition matrix to describe the evolution of the obstacle state over time, predict multiple future time steps through the Kalman filter covariance matrix, generate the future motion trajectory of the obstacle, and select the corresponding obstacle avoidance strategy according to the results of the previous classification model.

8. An obstacle perception system for lidar data analysis according to any one of claims 1 to 7, characterized in that: A data preprocessing module, a data fusion module, a feature extraction module, a classification model module, and a filter module. Specifically, The preprocessing module identifies and removes random noise through dynamic noise filtering, unifies the data reference framework through coordinate system transformation, and reduces the data volume and improves efficiency through point cloud compression technology; The data fusion module solves the sensor movement difference through multi-temporal frame spatial alignment, and enhances the information expression by fusing the feature vectors of different data sources at the feature level; The feature extraction module calculates the FPFH descriptor to capture the geometric characteristics of the point cloud, analyzes the reflection intensity to provide clues about the surface properties of the object, and helps to distinguish different objects; The classification model module divides the point cloud by PointNet++ to learn local to global features, and adaptively adjusts the weights through dynamic convolution to improve the adaptability to diverse 3D cloud data; The filter module initializes the state vector using the improved Kalman filter and accurately estimates the system motion state through the prediction and update loop.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the obstacle perception method based on lidar data analysis according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the obstacle perception method based on lidar data analysis according to any one of claims 1 to 7.

Images