Smoke identification method and device based on laser point cloud, equipment and storage medium

Abstract

The invention relates to the technical field of automatic driving, and discloses a laser point cloud smoke identification method, device and equipment and a storage medium, and the method comprises the steps: receiving a plurality of point cloud data detected by a laser radar when the current frame scans the surrounding environment, and determining the intensity value of each laser point cloud based on the comparison of the intensity value corresponding to each laser point cloud and an intensity threshold value; screening to obtain a first point cloud set, obtaining an initial density clustering model, configuring hyper-parameters in the initial density clustering model to generate an advanced density clustering model, processing the laser point clouds in the first point cloud set through the advanced density clustering model, and outputting a second point cloud set, and identifying each point cloud data clustered by the plurality of laser point clouds in the second point cloud set as smoke point clouds. The method does not depend on specific smoke shape or color features, and compared with methods based on perception fusion, multiple echoes and the like, the method is wider in applicability, is not affected by laser radar types, and can be suitable for smoke detection in various complex scenes.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving technology, and in particular to a laser point cloud smoke recognition method, device, equipment and storage medium. Background Art

[0002] LiDAR is a very important sensor type in the autonomous driving perception system. It has powerful environmental perception and processing capabilities. Its working principle is: it sends a detection signal to the target. After encountering an object, the detection signal will be reflected back and received by the LiDAR sensor, thereby achieving the purpose of detecting the surrounding environment. The results of LiDAR's detection of the surrounding environment can be represented in the form of a point cloud. The point cloud mainly includes: three-dimensional coordinates and reflection intensity in the environment. The reflection intensity is related to information such as the surface material and roughness of the object.

[0003] In the process of using LiDAR to identify the surrounding point cloud, if smoke is encountered, such as smoke and fumes emitted by chemical plants or incineration plants, the recognition ability is poor, and smoke and fumes and other media are easily identified as obstacles, resulting in the inability to accurately control the vehicle during the autonomous driving process, and even causing the autonomous driving system to fail to operate normally. Summary of the invention

[0004] In view of this, the present invention provides a laser point cloud smoke recognition method, device, equipment and storage medium to improve the smoke filtering capability in the field of robots and autonomous driving, thereby ensuring the normal operation of the autonomous driving system.

[0005] In a first aspect, the present invention provides a smoke recognition method using a laser point cloud, the method comprising:

[0006] Receive multiple point cloud data detected by the laser radar when scanning the surrounding environment in the current frame, each point cloud data includes an intensity value corresponding to the laser point cloud;

[0007] Based on the comparison between the intensity value corresponding to each laser point cloud and the intensity threshold, a first point cloud set is obtained by screening, wherein the intensity value of each laser point cloud in the first point cloud set is less than the intensity threshold;

[0008] Acquire an initial density clustering model, configure hyperparameters in the initial density clustering model, and generate an advanced density clustering model, wherein the advanced density clustering model is used to identify and cluster point cloud data;

[0009] The laser point clouds in the first point cloud set are processed by the advanced density clustering model to output a second point cloud set, and each point cloud data of a plurality of laser point clouds in the second point cloud set that are clustered into clusters is identified as a smoke point cloud.

[0010] The smoke recognition method of laser point cloud provided by the present invention firstly performs preliminary screening of the original point cloud data by introducing an intensity threshold, thereby excluding non-smoke point clouds with higher intensity, effectively reducing the amount of data for subsequent processing, and improving the pertinence of smoke recognition, thereby improving the overall recognition accuracy. Then, an advanced density clustering model is used to perform cluster analysis on the screened point cloud data, which can adaptively identify smoke clusters of different shapes, densities, and distribution characteristics, thereby enhancing the adaptability and robustness of the method under different environmental conditions.

[0011] Since the advanced density clustering model is generated by pre-configuring the hyperparameters of the initial density clustering model, clustering laser point clouds through the advanced density clustering model is more efficient, and smoke point clouds can be quickly and accurately identified from a large amount of point cloud data, shortening the processing time and improving real-time performance.

[0012] In addition, this method does not rely on specific smoke shapes or color features, but is based on the density distribution characteristics of point clouds for identification. Compared with methods based on perception fusion, multiple echoes, etc., this method is more widely applicable. It is not affected by the type of lidar and can be applied to smoke detection in a variety of complex scenarios.

[0013] In combination with the first aspect, in a possible implementation, obtaining an initial density clustering model, configuring hyperparameters in the initial density clustering model, and generating an advanced density clustering model include:

[0014] Obtaining an initial density clustering model, wherein the initial density clustering model includes hyperparameter constraints: distance metric information of the point cloud and a minimum number of samples;

[0015] Based on different configurations of the distance metric information and the minimum number of samples, determine a target hyperparameter in the test results of at least two sets of hyperparameters, wherein the target hyperparameter corresponds to a target distance metric information and a target minimum number of samples;

[0016] The target distance metric information and the target minimum number of samples in the target hyperparameters are configured in the initial density clustering model to generate the advanced density clustering model.

[0017] In combination with the first aspect, in another possible implementation, based on different configurations of the distance metric information and the minimum number of samples, determining a target hyperparameter from test results of at least two groups of hyperparameters includes:

[0018] Based on different configurations of the distance metric information and the minimum number of samples, calculating the precision and recall under each set of hyperparameter configurations;

[0019] Determine the evaluation index F1 corresponding to each set of hyperparameter configurations according to the precision and the recall;

[0020] The evaluation index F1 corresponding to each set of hyperparameter configurations is compared with the preset true value index, the evaluation index F1 closest to the preset true value index is determined, and the hyperparameter corresponding to the closest evaluation index F1 is used as the target hyperparameter.

[0021] In combination with the first aspect, in another possible implementation, determining the evaluation index F1 corresponding to each group of hyperparameter configurations according to the precision and the recall includes:

[0022] According to the precision and the recall, the evaluation index F1 is calculated using a first relational expression, where the first relational expression is:

[0023]

[0024] Among them, a represents precision, b represents recall, and F1 represents the evaluation index.

[0025] In combination with the first aspect, in another possible implementation, after processing the laser point cloud in the first point cloud set by the advanced density clustering model and outputting the second point cloud set, the method further includes:

[0026] Acquire a plurality of point cloud data detected by the laser radar when scanning the surrounding environment in the previous frame;

[0027] Get the 3D object detection model;

[0028] The plurality of point cloud data detected by the current frame scanning and the plurality of point cloud data detected by the previous frame scanning are respectively subjected to smoke recognition by the 3D object detection model to obtain a first recognition result and a second recognition result, wherein the first recognition result and the second recognition result at least include one or more irregular polygonal objects;

[0029] respectively calculating the area of ​​at least one irregular polygonal object in the first recognition result and the area of ​​at least one irregular polygonal object in the second recognition result;

[0030] Comparing the area differences of the irregular polygonal objects in the first recognition result and the irregular polygonal objects in the second recognition result one by one, and calculating the intersection-and-union ratio of the areas of the irregular polygonal objects in the previous frame and the current frame;

[0031] If the calculated intersection-and-union ratio is less than a threshold value, determining to use the laser point cloud on the area of ​​the irregular polygonal object corresponding to the intersection-and-union ratio as a third point cloud set;

[0032] The laser point cloud in the third point cloud set and the laser point cloud in the second point cloud set are intersected to obtain at least one intersecting laser point cloud, and the at least one intersecting laser point cloud is determined to be a smoke point cloud.

[0033] In this implementation, smoke is filtered based on the IOU information between frames of the 3D target detection model perception result, the laser point cloud intensity and density information, that is, the smoke point cloud is filtered using the laser point cloud intensity and density information, and the smoke point cloud is screened using the 3D target detection results between frames. The two methods are combined to take the intersection and finally filter out the smoke point cloud. This method can effectively filter the smoke information in the environment, reduce the probability of misidentification of smoke, and improve the accuracy of smoke point cloud recognition and filtering.

[0034] In addition, this method can be effectively applied to autonomous driving perception applications based on lidar as a sensor and has high practical value.

[0035] In combination with the first aspect, in another possible implementation, taking an intersection of the laser point cloud in the third point cloud set and the laser point cloud in the second point cloud set to obtain at least one intersecting laser point cloud, and determining that the at least one intersecting laser point cloud is a smoke point cloud, including:

[0036] Respectively obtaining intensity values ​​corresponding to each laser point cloud in the first recognition result and the second recognition result;

[0037] A smoke point cloud index obtained by filtering according to the intensity value and density of the point cloud, wherein the smoke point cloud index is represented by i, i=0, 1, ...n;

[0038] Obtain the corresponding smoke point cloud index according to the intersection-and-union ratio of adjacent frames in target recognition, where the corresponding smoke point cloud index is represented by j, where j=3, 4, ...m;

[0039] Based on the smoke point cloud indexes i and j, an intersection operation is performed to obtain the index value of the target smoke point cloud. The index value of the target smoke point cloud is expressed as:

[0040] The smoke point cloud is determined according to the index value k of the target smoke point cloud.

[0041] In combination with the first aspect, in another possible implementation, based on the comparison between the intensity value corresponding to each laser point cloud and the intensity threshold, the first point cloud set is screened and obtained, including:

[0042] Determine whether the intensity value of each of the laser point clouds is less than the intensity threshold; if so, count all laser point clouds whose intensity values ​​are less than the intensity threshold and record the index values ​​corresponding to the laser point clouds to generate the first point cloud set.

[0043] In a second aspect, the present invention provides a laser point cloud-based smoke recognition device, the device comprising:

[0044] A receiving module is used to receive a plurality of point cloud data detected by the laser radar when scanning the surrounding environment in the current frame, each point cloud data includes point cloud coordinates and intensity values ​​corresponding to the laser point cloud;

[0045] A screening module, configured to screen and obtain a first point cloud set based on a comparison between an intensity value corresponding to each laser point cloud and an intensity threshold, wherein the intensity value of each laser point cloud in the first point cloud set is less than the intensity threshold;

[0046] A processing module, used to obtain an initial density clustering model, configure hyperparameters in the initial density clustering model, and generate an advanced density clustering model, wherein the advanced density clustering model is used to identify and cluster point cloud data;

[0047] The recognition module is used to process the laser point cloud in the first point cloud set through the advanced density clustering model, output a second point cloud set, and recognize that each point cloud data of multiple laser point clouds in the second point cloud set that are clustered into clusters is a smoke point cloud.

[0048] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the laser point cloud smoke recognition method of the first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0049] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the laser point cloud smoke identification method of the first aspect or any corresponding embodiment thereof.

[0050] In addition, the present invention provides a computer program product, including computer instructions, which are used to enable a computer to execute the laser point cloud smoke recognition method of the above-mentioned first aspect or any corresponding embodiment thereof.

[0051] The present invention provides a laser point cloud smoke recognition method, device, equipment and storage medium, which realizes the filtering of smoke in an environment containing smoke based on the IOU information between frames of the laser point cloud 3D perception result, the laser point cloud intensity and density information. Compared with the methods based on perception fusion, multiple echoes, etc., this method has a wider applicability and is not affected by the type of laser radar. It combines deep learning target detection and point cloud geometric information and density information, and solves the smoke filtering problem in a variety of information fusion methods. It can quickly and accurately identify smoke point clouds from a large amount of point cloud data, shorten the processing time, improve the efficiency of smoke point cloud recognition and filtering, and also has a wide and high practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0053] Figure 1 is a schematic diagram of a laser point cloud system according to an embodiment of the present invention;

[0054] Figure 2 is a schematic flow chart of a smoke recognition method using a laser point cloud according to an embodiment of the present invention;

[0055] Figure 3 is a flow chart of another smoke recognition method using laser point cloud according to an embodiment of the present invention;

[0056] Figure 4 is a schematic diagram of a process of generating an advanced density clustering model according to an embodiment of the present invention;

[0057] Figure 5 is a flow chart of another smoke recognition method using laser point cloud according to an embodiment of the present invention;

[0058] Figure 6 is a flow chart of another smoke recognition method using laser point cloud according to an embodiment of the present invention;

[0059] Figure 7 is a structural block diagram of a smoke recognition device based on a laser point cloud according to an embodiment of the present invention;

[0060] Figure 8 It is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0061] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0062] First, the application scenarios and related technical terms of the technical solutions provided in the embodiments of the present invention are described.

[0063] The technical solution of the embodiment of the present invention relates to the field of autonomous driving technology. Figure 1 , is a schematic diagram of a laser point cloud system according to an embodiment of the present invention. Figure 1 As shown in the figure, some factories (such as chemical plants, incineration plants) will emit some smoke, which will cause smoke and pollution on some roads inside the factories. Therefore, vehicles with autonomous driving capabilities are needed to pass through these roads to achieve the goals of transporting goods. Figure 1 As shown, a vehicle equipped with a lidar needs to pass through the internal route of a factory with waste smoke, so the vehicle needs to be able to effectively determine that there is smoke ahead to achieve autonomous driving capabilities.

[0064] An autonomous driving system is an intelligent system that can automatically sense and locate a vehicle without human intervention, and then complete route planning and control. The autonomous driving system is mainly divided into modules such as perception, positioning, planning, and control, among which the perception module is an important part of the autonomous driving system.

[0065] The perception system mainly refers to the vehicle's use of sensors to perceive and identify the environment, and can perform static obstacle recognition, traffic light recognition, and mobile obstacle detection and tracking. The perception system sensors are mainly divided into cameras, lidar, millimeter wave radar, and ultrasonic radar. Different sensors play different roles in the autonomous driving perception system, and different sensors need to be used for environmental perception in different actual scenarios.

[0066] LiDAR is a very important sensor type in the autonomous driving perception system. It has powerful environmental perception and processing capabilities. Its working principle is: it sends a detection signal to the target. After encountering an object, the signal will be reflected back and received by the LiDAR sensor, thereby achieving the goal of detecting the nearby environment. LiDAR is mainly divided into two types according to its structural characteristics: mechanical LiDAR and solid-state LiDAR. The results of LiDAR's perception of the surrounding environment can be represented by a point cloud. The point cloud mainly includes: the three-dimensional coordinate (XYZ) value and reflection intensity in the environment. The reflection intensity is related to the surface material, roughness and other information of the target.

[0067] Smoke recognition and filtering is a perception scenario in the autonomous driving system. LiDAR has poor recognition ability for media such as smoke and will identify smoke as an obstacle, which will cause the autonomous driving system to fail to operate normally in smoke scenarios. In order to solve the above problems, the main methods currently used include radar with multiple echoes, surface LiDAR, and perception fusion. However, the current methods have the following main defects:

[0068] First, it relies on other types of sensors to fuse the perception effects for processing. This method cannot operate normally when there is only one sensor.

[0069] Second, it mainly uses laser radars with different characteristics to solve the problem from the hardware level (multiple echoes, surface laser radar), and its scope of application is relatively small.

[0070] In order to solve the above problems, an embodiment of the present invention provides a laser point cloud smoke filtering method (or recognition method), which uses point cloud intensity and point cloud geometric characteristics to fuse results, so as to solve the above-mentioned problems such as the small scope of application and improve the smoke filtering capabilities in the field of robots and autonomous driving.

[0071] The embodiments of the present invention provide various method embodiments. It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0072] In this embodiment, a smoke recognition method using laser point cloud is provided, which can be applied to a vehicle controller, such as ECU (Electronic Control Unit), also known as "on-board computer" or "on-board computer", which is a microcomputer controller designed for automobiles. In addition, it can also be other control circuits or computer equipment, computing platforms, etc., which are not limited in this embodiment.

[0073] Furthermore, the computing platform mentioned above may be a simulation framework with AWSIM (Autoware Simulation) function. AWSIM is a powerful, flexible and scalable autonomous driving simulation framework, which provides important support and promotion for the development of autonomous driving technology. The autonomous driving simulation framework can automatically simulate and generate various types of data. The tool is used to generate smoke data, mainly in road tests, and the data is input into the Autoware.universe framework as real laser point cloud data. This part of the data is mainly test data, which is used to test the final algorithm effect.

[0074] In addition to the above test data, AWSIM is required to generate labeled data, which annotates non-smoke point clouds and smoke point clouds. The labeled data is used for hyperparameter selection and verification in the following steps.

[0075] like Figure 2 FIG. 1 is a flow chart of a smoke recognition method using a laser point cloud according to an embodiment of the present invention. The method includes:

[0076] Step S101: receiving a plurality of point cloud data detected by a laser radar when scanning the surrounding environment in a current frame, wherein each point cloud data includes an intensity value corresponding to the laser point cloud.

[0077] The laser radar can be installed on the vehicle to detect the surrounding environment in real time as the vehicle travels. The laser radar emits laser detection signals to the surroundings, such as a 360° range, and feeds back the signals, which can be represented by laser point cloud data. The laser radar sends all the collected point cloud data to the vehicle controller or computing platform. This embodiment takes the vehicle controller as an example of the receiving device.

[0078] Among them, laser point cloud data is the data transmitted back to the laser radar after the laser beam is emitted to the target object by the laser radar, and it is composed of many points. In the laser point cloud data, each point cloud contains information such as three-dimensional coordinates (XYZ) and laser reflection intensity (Intensity). This information together constitutes a detailed description of the scanning environment and provides a basis for subsequent point cloud processing, analysis and application. For example, in the field of autonomous driving, the intensity information of laser point clouds can be used to identify road markings, obstacle detection, pedestrian recognition and other tasks to improve the perception and safety of autonomous driving systems.

[0079] In this embodiment, each laser point cloud is composed of (x, y, z, p). Among them, (x, y, z) is the coordinate information of a certain point cloud, indicating the specific position in three-dimensional space (XYZ axis), and p is the intensity of the laser point cloud. The intensity corresponding to the point cloud emitted and returned by the laser point cloud mainly refers to the reflection intensity value returned by each point when the laser radar sensor scans the environment. These intensity values ​​represent the energy returned by the laser pulse, which is usually closely related to the reflection characteristics of the surface of the object. For example, the intensity value is affected by factors such as the material and roughness of the object surface, the laser incident angle, the scanning distance, the atmospheric conditions (such as smoke, haze, dust, etc.), the laser wavelength and the energy density. The intensity value is usually within a relative range, such as 0 to 255. For the point cloud intensity of smoke, the range of its values ​​is somewhat different from that of other objects. Therefore, according to the intensity information in the point cloud result, the point cloud within a certain range can be selected and the index value information of its point cloud can be recorded.

[0080] In addition, each detection by the laser radar can be called a scan of a frame. The laser radar scans and detects periodically, and obtains a detection result in each frame. Each detection result can collect multiple point cloud data. Specifically, the number of point cloud data can be dozens, or hundreds or thousands.

[0081] Step S102: Based on the comparison between the intensity value corresponding to each laser point cloud and the intensity threshold, a first point cloud set is screened and obtained.

[0082] The intensity value of each laser point cloud in the first point cloud set is less than the intensity threshold. Specifically, one implementation is to screen and obtain the first point cloud set based on the comparison between the intensity value corresponding to each laser point cloud and the intensity threshold, including: determining whether the intensity value of each laser point cloud is less than the intensity threshold; if so, counting all laser point clouds whose intensity values ​​are less than the intensity threshold and recording the index value corresponding to the laser point cloud to generate the first point cloud set.

[0083] The formula is as follows:

[0084] p(i)<P

[0085] In the above formula, p(i)i=0,1,…n represents the intensity of each laser point cloud, P is the intensity threshold of the laser point cloud, when the intensity value of a point cloud is greater than or equal to the intensity threshold, the point cloud is considered to be a normal point cloud. Otherwise, the point cloud is considered to be an abnormal point cloud, and the index value i of the abnormal point cloud is recorded. Figure 3 As shown, this step can also be understood as sorting the point cloud intensities, and summarizing all point clouds with intensities less than the intensity threshold P into a point cloud set, namely, the first point cloud set.

[0086] It should be noted that, in order to filter out the point cloud in this embodiment, the intensity threshold P can be set to a larger threshold, for example, P=30, so as to ensure that all abnormal point clouds are classified into the first point cloud set, so as to facilitate further secondary screening of all point cloud data in the first point cloud set.

[0087] Step S103: obtaining an initial density clustering model, configuring hyperparameters in the initial density clustering model, and generating an advanced density clustering model.

[0088] Among them, the advanced density clustering model is used to identify and cluster point cloud data. The initial density clustering model is a spatial clustering algorithm, such as the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) module or the DBSCAN algorithm. Furthermore, DBSCAN is a density-based spatial clustering algorithm that can divide areas with sufficiently high density into clusters and can find clusters of arbitrary shapes in noisy data sets. Unlike distance-based clustering algorithms such as K-means, the DBSCAN algorithm does not require the number of clusters to be specified in advance, and can identify and process noise points.

[0089] In an algorithmic model, hyperparameters are parameters whose values ​​are set before the learning process begins, rather than parameters learned through data during training. Hyperparameters are often used to control the learning process of the model and affect the final performance and performance of the model. Their selection has a crucial impact on the performance of the model, but choosing the right hyperparameters is often a challenge because the optimal value of the hyperparameter depends on the specific problem and data set. In this embodiment, there is a specific implementation method for the correct selection of hyperparameters, which is specifically referred to in the following embodiments.

[0090] Optionally, in this embodiment, the finally selected hyperparameter is referred to as a target hyperparameter, which specifically includes target distance measurement information and a target minimum number of samples of the point cloud.

[0091] This step configures the target distance measurement information and the target minimum sample number in the target hyperparameters into the initial DBSCAN algorithm to generate a new DBSCAN algorithm, which is called an advanced density clustering model. The advanced density clustering model is used to provide algorithm support for subsequent laser point cloud data recognition and filtering.

[0092] Step S104: Processing the laser point clouds in the first point cloud set through the advanced density clustering model, outputting a second point cloud set, and identifying each point cloud data of the plurality of laser point clouds in the second point cloud set that are clustered into clusters as smoke point clouds.

[0093] like Figure 3 As shown, in this step, an advanced density clustering model is used to cluster all point clouds in the first point cloud set, and the point clouds are clustered into clusters, which is also called "clustering and de-anomaly" operation, that is, the points with target distance measurement information and target minimum sample number are clustered into a cluster, and the point clouds belonging to a cluster are non-outlier points; and the point clouds not belonging to a cluster are discrete point clouds or outlier points; in this embodiment, it is set that the point clouds belonging to the same cluster (non-outlier point clouds) form a smoke point cloud; and the discrete points or outlier point clouds belong to normal point clouds.

[0094] After the above step S104 identification, multiple clusters can be obtained, each cluster contains the target number of samples and the range is within the target distance measurement information (for example, the target distance is within the radius d), and the point cloud set formed by all clustered clusters is the above second point cloud set.

[0095] The smoke recognition method of laser point cloud provided by the present invention firstly performs preliminary screening of the original point cloud data by introducing an intensity threshold, thereby excluding non-smoke point clouds with higher intensity, effectively reducing the amount of data for subsequent processing, and improving the pertinence of smoke recognition, thereby improving the overall recognition accuracy. Then, an advanced density clustering model is used to perform cluster analysis on the screened point cloud data, which can adaptively identify smoke clusters of different shapes, densities, and distribution characteristics, thereby enhancing the adaptability and robustness of the method under different environmental conditions.

[0096] Since the advanced density clustering model is generated by pre-configuring the hyperparameters of the initial density clustering model, clustering laser point clouds through the advanced density clustering model is more efficient, and smoke point clouds can be quickly and accurately identified from a large amount of point cloud data, shortening the processing time and improving real-time performance.

[0097] In addition, this method does not rely on specific smoke shapes or color features, but is based on the density distribution characteristics of point clouds for identification. Compared with methods based on perception fusion, multiple echoes, etc., this method is more widely applicable. It is not affected by the type of lidar and can be applied to smoke detection in a variety of complex scenarios.

[0098] It should be noted that the method provided in this embodiment can also be applied to fields related to autonomous driving through a high-performance, low-power computing platform. For example, a computing platform that integrates the Autoware.Universe functional module. Autoware.Universe is an autonomous driving framework that has perception, positioning, planning and control modules for autonomous driving, and can achieve the entire autonomous driving from perception input to control output. The computing platform used in this embodiment includes but is not limited to Autoware.Universe.

[0099] Furthermore, the CPU, GPU, operating system, ROS2 version and software framework carried by the computing platform can be customized by the system or the user, and this embodiment does not limit this.

[0100] like Figure 4 As shown, the above step S103, obtaining an initial density clustering model, configuring the hyperparameters in the initial density clustering model, and generating an advanced density clustering model, specifically includes:

[0101] Step S103 - 1 : obtaining an initial density clustering model, wherein the initial density clustering model includes hyperparameter constraints: distance measurement information of the point cloud and a minimum number of samples.

[0102] The distance measurement information of the point cloud refers to an area with a radius of d centered on a certain point cloud position, which is a parameter used to set the size of the point cloud range. The minimum number of samples refers to the number of all point clouds contained in the area with a radius of d. The minimum value of the number of point clouds is the minimum number of samples. For example, the minimum number of samples is 100, and the scanning radius distance d = 50cm, which means that within the range of the point cloud center (x1, y1, z1) coordinates and the scanning radius d of 50cm, the number of other point clouds contained is at least 100 clusters.

[0103] Optionally, the initial density clustering model is a density clustering DBSCAN algorithm.

[0104] Step S103 - 2 : Based on different configurations of distance metric information and minimum number of samples, determine target hyperparameters from test results of at least two groups of hyperparameters.

[0105] Based on different distance measurement information (such as radius distance d) and minimum number of samples, different groups of hyperparameters can be set. One radius distance d and one minimum number of samples are a set of hyperparameters. Each set of hyperparameters and the aforementioned "generate labeled data through AWSIM, which annotates non-smoke point clouds and smoke point clouds, and use the labeled data for hyperparameter selection and verification in the following steps" can calculate an experimental result, and determine one set as the target hyperparameter from the experimental results of at least two sets of hyperparameters.

[0106] The test result can be determined by comparing the evaluation index F1 with the true value. The target hyperparameter corresponds to a target distance measurement information and a target minimum number of samples.

[0107] Step S103 - 3 : configuring the target distance metric information in the target hyperparameters and the target minimum number of samples in the initial density clustering model to generate an advanced density clustering model.

[0108] The method of setting a threshold based on the commonly used manual experience to determine whether it is smoke has a low accuracy rate. If the intensity threshold is set high in step S102, smoke filtering is performed based on the point cloud intensity, then part of the point cloud will be classified as a smoke point cloud. If the intensity threshold is set low, smoke point clouds will be classified as normal point clouds, which will eventually cause missed detection or false detection. Therefore, the first point cloud set needs to be further processed. This step uses an advanced density clustering model to perform a secondary screening on the first point cloud set generated by the first screening based on the geometric information of the laser point cloud to improve the accuracy.

[0109] In this embodiment, an advanced density clustering model is used to cluster the laser point clouds in the first point cloud set. The premise of this method is that the point cloud density in the smoke is relatively high. For point clouds with lower density, the possibility that they are smoke point clouds is lower, and they are classified as normal point clouds. The clustered point clouds have higher density and can be identified as smoke point clouds.

[0110] Furthermore, the above step S103-2: determining the target hyperparameter in the test results of at least two groups of hyperparameters based on different configurations of the distance metric information and the minimum number of samples, specifically includes:

[0111] Based on different configurations of the distance metric information and the minimum number of samples, the precision and recall under each set of hyperparameter configurations are calculated; based on the precision and the recall, the evaluation index F1 corresponding to each set of hyperparameter configurations is determined; the evaluation index F1 corresponding to each set of hyperparameter configurations is compared with the preset true value index, the evaluation index F1 closest to the preset true value index is determined, and the hyperparameter corresponding to the closest evaluation index F1 is used as the target hyperparameter.

[0112] Precision: The number of samples predicted to be positive and actually positive divided by the total number of samples predicted to be positive, that is, the precision is calculated by true positives (TP) and false positives (FP). Recall: The number of samples predicted to be positive and actually positive divided by the total number of samples actually positive, that is, the recall is calculated by TP and false negatives (FN). Among them, positive and negative are two sides of the binary classification problem.

[0113] The above closest can be understood as: the difference between the evaluation index F1 calculated by the target hyperparameter and the preset true value index (such as F0) is the smallest among all evaluation indicators. For example, when the difference is 0, it means that the target evaluation index F1 calculated by the current target hyperparameter is equal to F0.

[0114] According to the precision and the recall, determining the evaluation index F1 corresponding to each set of hyperparameter configurations includes: calculating the evaluation index F1 according to the precision and the recall using a first relational expression, the first relational expression being:

[0115]

[0116] Among them, a represents precision, b represents recall, and F1 represents the evaluation index.

[0117] One way to obtain the above-mentioned preset truth value index is to collect laser point cloud data sets for manual annotation. Relying on manual experience, the smoke information in the data set can be distinguished and annotated, and the annotated data set can be used to calculate the preset truth value index (F0). Based on the different configurations of the distance metric information and the minimum number of samples, it is assumed that a total of 10 sets of hyperparameters are configured, each set of hyperparameters corresponds to a distance and a minimum number of samples; each set of hyperparameters can calculate a precision rate a and a recall rate b, and calculate an evaluation index F1 according to the above-mentioned first relationship.

[0118] For example, assuming that a total of 10 sets of hyperparameters are set, 10 evaluation indicators F1 can be calculated accordingly, such as evaluation indicators F1-1 to F1-10, and each evaluation indicator is compared with the preset true value indicator F0. A set of evaluation indicators F1-x that is equal to or closest to the true value indicator F0 is used as the target hyperparameter. For example, the second set of hyperparameters F1-1 is closest to F0, so F1-2 is determined as the target hyperparameter. At this time, the distance d2 and a minimum number of samples n2 in F1-2 are the target distance measurement information and the target minimum number of samples.

[0119] In another embodiment, in addition to the above process, the filtering effect of the smoke point cloud can be further improved based on the IOU information between different frames in the point cloud 3D object detection result. Figure 5 FIG. 2 is a flow chart of another method for smoke recognition based on laser point cloud according to an embodiment of the present invention. In addition to the laser point cloud intensity screening and laser point cloud cluster screening in the above embodiment, the process also includes: screening based on 3D target detection and inter-frame IOU calculation to obtain a third point cloud set, and then combining the second point cloud output by the laser point cloud cluster screening with the third point cloud set to obtain the intersection, that is, filtering the point cloud smoke again, and finally obtaining a filtered smoke point cloud.

[0120] Specific process, such as Figure 6 As shown, the above step S104, after processing the laser point cloud in the first point cloud set by the advanced density clustering model and outputting the second point cloud set, further includes:

[0121] Step S201: Acquire a plurality of point cloud data detected by the laser radar when scanning the surrounding environment in the previous frame.

[0122] The vehicle controller periodically obtains point cloud data from the laser radar in each sampling period (sampling frame). For example, in the above step S101, multiple point cloud data of the current frame are obtained. In this embodiment, the sampling point cloud data of the adjacent frames are also obtained during the execution of the "3D object detection" process. This embodiment is explained by taking the previous frame of the current frame as an example.

[0123] Step S202: Acquire a 3D object detection model.

[0124] Among them, the 3D object detection model is used to identify 3D objects in point cloud data.

[0125] Optionally, the 3D target detection model is the CenterPoint model. The CenterPoint model is a 3D target detection algorithm based on lidar point cloud, which is an Anchor-Free 3D object detector. The main function of the CenterPoint model is to provide real-time, accurate and stable 3D target detection effects in complex urban road scenes. The CenterPoint model has a wide range of applications in the field of autonomous driving, especially in the Apollo autonomous driving platform. The CenterPoint model has been optimized to improve detection accuracy and recall rate. The characteristics of this model are: taking the point cloud as input, the center point of the 3D object under Bird-View as the key point, and regressing the size, direction and speed of the object based on the key point detection method. It has the advantages of real-time, accuracy, stability and ease of use.

[0126] Step S203: The plurality of point cloud data detected by the current frame scan and the plurality of point cloud data detected by the previous frame scan are respectively subjected to smoke recognition by the 3D target detection model to obtain a first recognition result and a second recognition result.

[0127] The first recognition result and the second recognition result at least include one or more irregular polygonal objects. Specifically, after the multiple point cloud data detected by the current frame scanning are recognized by the 3D target detection model, the first recognition result is output; after the multiple point cloud data detected by the previous frame scanning are recognized by the 3D target detection model, the second recognition result is output.

[0128] In this embodiment, the point cloud data is input into a deep learning model (such as a CenterPoint model) for target detection. The deep learning model can be used to detect smoke, mainly by statistically analyzing the category and shape information in the target results. The recognition result category of smoke is generally unknown (understood as only the category is recognized, but the specific object is unknown). In this embodiment, unknown is understood as the recognition result is usually an irregular polygonal object.

[0129] Step S204: Calculate the area of ​​at least one irregular polygonal object in the first recognition result and the area of ​​at least one irregular polygonal object in the second recognition result respectively.

[0130] The area of ​​the irregular polygonal object may be calculated by using the point cloud coordinates of multiple point clouds. In addition, the area of ​​the irregular polygonal object may also be calculated by other methods, which are not limited in this embodiment.

[0131] Step S205: comparing the area differences of the irregular polygonal objects in the first recognition result and the irregular polygonal objects in the second recognition result one by one, and calculating the intersection-over-union ratio of the areas of the irregular polygonal objects in the previous frame and the current frame.

[0132] Target tracking refers to establishing instance correspondences between different frames. In a smoky environment, the shapes of the recognition results detected at different times are usually very different due to the real-time changes of the smoke. Therefore, the point cloud recognition result of the current frame (the first recognition result) is compared with the recognition result of the previous frame (the second recognition result), mainly comparing the IOU values ​​(Intersection over Union) of the polygonal objects corresponding to the two adjacent frames. When the IOU value is small, it indicates that the shape of the irregular object is very different and the probability of smoke is high.

[0133] Assume S t is the area of ​​the polygonal object in the previous frame, S t+1 is the area of ​​the polygonal object in the current frame. The intersection and ratio of the areas of the two polygons are taken as the final IOU value. When the IOU value is less than the threshold, the corresponding point cloud is considered to be a smoke point cloud. A method for calculating the IOU value is shown in the second relation below:

[0134]

[0135] Step S206: If the calculated intersection-and-union ratio is less than a threshold value, the laser point cloud on the area of ​​the irregular polygonal object corresponding to the intersection-and-union ratio is determined as a third point cloud set.

[0136] Step S207: intersecting the laser point cloud in the third point cloud set with the laser point cloud in the second point cloud set to obtain at least one intersecting laser point cloud, and determining that the at least one intersecting laser point cloud is a smoke point cloud.

[0137] For example, in the aforementioned embodiment, after steps S101 to S104, the second point cloud set contains 500 laser point clouds, which are identified as smoke point clouds; after steps S201 to S206, the third point cloud set contains 600 laser point clouds, which are also identified as being smoke point clouds with a high probability; then the 500 laser point clouds in the second point cloud set and the 600 laser point clouds in the third point cloud set are intersected to find common laser point clouds, for example, the intersection result is 300 laser point clouds, and then the 300 intersected laser point clouds are determined to be accurately identified smoke point clouds, and the 300 smoke point clouds are filtered out from the entire frame image, or subjected to subsequent processing, etc.

[0138] In this implementation, smoke is filtered based on the IOU information between frames of the 3D target detection model perception result, the laser point cloud intensity and density information, that is, the smoke point cloud is filtered using the laser point cloud intensity and density information, and the smoke point cloud is screened using the 3D target detection results between frames. The two methods are combined to take the intersection and finally filter out the smoke point cloud. This method can effectively filter the smoke information in the environment, reduce the probability of misidentification of smoke, and improve the accuracy of smoke point cloud recognition and filtering.

[0139] In addition, this method can be effectively applied to autonomous driving perception applications based on lidar as a sensor and has high practical value.

[0140] Furthermore, the above step S207, intersecting the laser point cloud in the third point cloud set with the laser point cloud in the second point cloud set to obtain at least one intersecting laser point cloud, and determining that the at least one intersecting laser point cloud is a smoke point cloud, specifically includes:

[0141] The intensity values ​​corresponding to the laser point clouds in the first recognition result and the second recognition result are obtained respectively; the smoke point cloud index is obtained by filtering according to the intensity value and density of the point cloud, and the smoke point cloud index is represented by i, i=0,1,…n; the corresponding smoke point cloud index is obtained according to the intersection-and-union ratio of adjacent frames in the target recognition, and the corresponding smoke point cloud index is represented by j, j=3,4,…m; based on the smoke point cloud indexes i and j, an intersection operation is performed to obtain the index value of the target smoke point cloud, and the index value of the target smoke point cloud is represented by k as follows: k=i∩j.

[0142] The smoke point cloud is determined according to the index value k of the target smoke point cloud.

[0143] According to the above steps, the smoke point cloud index i=0,1,…n is obtained by filtering according to the point cloud intensity and density, and the corresponding smoke point cloud index j=3,4,…m is obtained according to the IOU value of the adjacent frames in the target recognition. Based on the above two point cloud index values, the intersection operation is performed to obtain the final smoke point cloud index value, that is, the final point cloud index value k. The point cloud index value k can be expressed as k=i∩j.

[0144] According to the index value, the smoke is filtered to finally achieve the goal of smoke filtering.

[0145] The present embodiment provides a method for smoke recognition based on laser point cloud, which realizes filtering smoke in an environment containing smoke based on IOU information between frames of laser point cloud 3D perception results, laser point cloud intensity and density information. Compared with methods based on perception fusion, multiple echoes, etc., this method has a wider range of applicability and is not affected by the type of laser radar. It combines deep learning target detection and point cloud geometric information and density information, and solves the smoke filtering problem in a variety of information fusion methods. It can quickly and accurately identify smoke point clouds from a large amount of point cloud data, shorten processing time, improve smoke point cloud recognition and filtering efficiency, and has a wide and high practical value.

[0146] In this embodiment, a laser point cloud-based smoke identification device is also provided, which is used to implement the laser point cloud-based smoke identification method described in the above embodiment, which has been described and will not be repeated. As used below, the term "module" can implement a combination of software and / or hardware for a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0147] This embodiment provides a smoke recognition device based on laser point cloud, such as Figure 7 As shown, the device includes: a receiving module 710, a screening module 720, a processing module 730 and an identification module 740. In addition, the device may also include other more or fewer modules / units, such as a storage module, a sending module, etc., which is not limited in this embodiment.

[0148] Among them, the receiving module 710 is used to receive multiple point cloud data detected by the laser radar when scanning the surrounding environment in the current frame, and each point cloud data includes point cloud coordinates and intensity values ​​corresponding to the laser point cloud.

[0149] The screening module 720 is used to screen and obtain a first point cloud set based on a comparison between the intensity value corresponding to each laser point cloud and the intensity threshold, wherein the intensity value of each laser point cloud in the first point cloud set is less than the intensity threshold.

[0150] The processing module 730 is used to obtain an initial density clustering model, configure hyperparameters in the initial density clustering model, and generate an advanced density clustering model, wherein the advanced density clustering model is used to identify and cluster point cloud data.

[0151] The identification module 740 is used to process the laser point clouds in the first point cloud set through the advanced density clustering model, output a second point cloud set, and identify each point cloud data of multiple laser point clouds in the second point cloud set that are clustered into clusters as smoke point clouds.

[0152] In a specific implementation, the receiving module 710 is further used to obtain a plurality of point cloud data detected by the laser radar when scanning the surrounding environment in the previous frame; and to obtain a 3D target detection model.

[0153] The screening module 720 is also used to perform smoke recognition on the multiple point cloud data detected by the current frame scan and the multiple point cloud data detected by the previous frame scan, respectively, through the 3D target detection model to obtain a first recognition result and a second recognition result, wherein the first recognition result and the second recognition result include at least one or more irregular polygonal objects.

[0154] The processing module 730 is also used to respectively calculate the area of ​​at least one irregular polygonal object in the first recognition result and the area of ​​at least one irregular polygonal object in the second recognition result; compare the difference between the area of ​​the irregular polygonal object in the first recognition result and the area of ​​the irregular polygonal object in the second recognition result one by one, and calculate the intersection-and-union ratio of the areas between the irregular polygonal objects in the previous frame and the current frame; and, if the calculated intersection-and-union ratio is less than a threshold value, determine to use the laser point cloud on the area of ​​the irregular polygonal object corresponding to the intersection-and-union ratio as the third point cloud set.

[0155] The processing module 730 is further configured to intersect the laser point cloud in the third point cloud set with the laser point cloud in the second point cloud set to obtain at least one intersecting laser point cloud.

[0156] The identification module 740 is further configured to determine that at least one of the intersecting laser point clouds is a smoke point cloud.

[0157] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0158] The laser point cloud-based smoke recognition device in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0159] The present invention provides a laser point cloud smoke identification device, which can filter smoke in a smoke-containing environment based on the IOU information between laser point cloud 3D perception result frames, laser point cloud intensity and density information. Compared with the methods based on perception fusion, multiple echoes, etc., this method has a wider applicability and is not affected by the type of laser radar. It combines deep learning target detection and point cloud geometric information and density information, and solves the smoke filtering problem in a variety of information fusion methods. It can quickly and accurately identify smoke point clouds from a large amount of point cloud data, shorten the processing time, improve the efficiency of smoke point cloud recognition and filtering, and has a wide and high practical value.

[0160] The embodiment of the present invention also provides a computer device having the above Figure 7 The smoke recognition device based on laser point cloud is shown.

[0161] See also Figure 8 , is a schematic diagram of the structure of a computer device provided by an optional embodiment of the present invention, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. The various components are connected to each other using different buses for communication, and can be installed on a common motherboard or in other ways as needed. The processor can process instructions executed in the computer device, including instructions stored in or on the memory to display graphical information of the GUI on an external input / output device (such as a display device coupled to the interface).

[0162] In some optional embodiments, if desired, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Multiple computer devices can also be connected, and each device provides part of the necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 8 A processor 10 is taken as an example.

[0163] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0164] The memory 20 stores instructions executable by at least one processor 10, so that the at least one processor 10 executes the smoke recognition method based on the laser point cloud shown in the above embodiment.

[0165] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0166] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 20 may also include a combination of the above types of memory.

[0167] The computer device also includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30 and the output device 40 may be connected via a bus or other means. Figure 8 The example of connecting through bus is taken in the following.

[0168] The input device 30 can receive input digital or character information, and generate key signal input related to the user settings and function control of the computer device, such as a touch screen, a keypad, a mouse, a track pad, a touch pad, an indicator bar, one or more mouse buttons, a trackball, a joystick, etc. The output device 40 may include a display device, an auxiliary lighting device (e.g., an LED) and a tactile feedback device (e.g., a vibration motor), etc. The above-mentioned display device includes but is not limited to a liquid crystal display, a light emitting diode, a display and a plasma display. In some optional embodiments, the display device can be a touch screen.

[0169] In addition, the computer device also includes at least one communication interface for the computer device to communicate with other devices or communication networks. Figure 8 At least one communication interface is not shown.

[0170] Optionally, the above-mentioned computer device may be a vehicle controller or a control circuit, such as an ECU.

[0171] In addition, this embodiment also provides a vehicle, which includes a vehicle controller, and the vehicle controller is used to execute the above Figures 2 to 6 The method flow shown solves the smoke filtering problem and can quickly and accurately identify smoke point clouds from a large amount of point cloud data, thereby shortening the processing time and improving the efficiency of smoke point cloud recognition and filtering.

[0172] In addition, the vehicle is also provided with laser radars, wherein the number of the laser radars can be set according to actual needs, and this embodiment does not impose any restrictions on this.

[0173] An embodiment of the present invention also provides a computer-readable storage medium, and the above-mentioned method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented by downloading through a network and originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware.

[0174] The storage medium may be a disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid state drive, etc.; further, the storage medium may also include a combination of the above-mentioned types of memories. It is understood that a computer, a processor, a microprocessor controller or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor or hardware, the smoke recognition method of the laser point cloud shown in the above embodiment is implemented.

[0175] Embodiments of the present application may also provide a computer program product, including computer program instructions, which, when executed by a processor, cause the processor to perform the steps in the above method. Wherein, the computer program product may be written in any combination of one or more programming languages ​​to perform program codes for performing the operations of the disclosed embodiments, wherein the programming languages ​​include object-oriented programming languages, such as Java, C++, etc., and also include conventional procedural programming languages, such as "C" language or similar programming languages. The program code may be executed entirely on a user computing device, partially on a user device, as an independent software package, partially on a user computing device, partially on a remote computing device, or entirely on a remote computing device or server.

[0176] The above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention, rather than to limit them. Although the embodiments of the present invention are described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some of the technical features therein by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A smoke recognition method based on laser point cloud, characterized in that: The method comprises: Receive multiple point cloud data detected by the laser radar when scanning the surrounding environment in the current frame, each point cloud data includes an intensity value corresponding to the laser point cloud; Based on the comparison between the intensity value corresponding to each laser point cloud and the intensity threshold, a first point cloud set is obtained by screening, wherein the intensity value of each laser point cloud in the first point cloud set is less than the intensity threshold; Acquire an initial density clustering model, configure hyperparameters in the initial density clustering model, and generate an advanced density clustering model, wherein the advanced density clustering model is used to identify and cluster point cloud data; The laser point clouds in the first point cloud set are processed by the advanced density clustering model to output a second point cloud set, and each point cloud data of a plurality of laser point clouds in the second point cloud set that are clustered into clusters is identified as a smoke point cloud.

2. The method according to claim 1, characterized in that Obtaining an initial density clustering model, configuring hyperparameters in the initial density clustering model, and generating an advanced density clustering model, including: Acquire an initial density clustering model, wherein the initial density clustering model includes hyperparameter constraints: distance metric information of the point cloud and a minimum number of samples; Based on different configurations of the distance metric information and the minimum number of samples, determine a target hyperparameter in the test results of at least two sets of hyperparameters, wherein the target hyperparameter corresponds to a target distance metric information and a target minimum number of samples; The target distance metric information and the target minimum number of samples in the target hyperparameters are configured in the initial density clustering model to generate the advanced density clustering model.

3. The method according to claim 2, characterized in that Based on different configurations of the distance metric information and the minimum number of samples, determining a target hyperparameter from test results of at least two sets of hyperparameters includes: Based on different configurations of the distance metric information and the minimum number of samples, calculating the precision and recall under each set of hyperparameter configurations; Determine the evaluation index F1 corresponding to each set of hyperparameter configurations according to the precision and the recall; The evaluation index F1 corresponding to each set of hyperparameter configurations is compared with the preset true value index, the evaluation index F1 closest to the preset true value index is determined, and the hyperparameter corresponding to the closest evaluation index F1 is used as the target hyperparameter.

4. The method according to claim 3, characterized in that Determining the evaluation index F1 corresponding to each set of hyperparameter configurations according to the precision and the recall includes: According to the precision and the recall, the evaluation index F1 is calculated using a first relational expression, where the first relational expression is: Among them, a represents precision, b represents recall, and F1 represents the evaluation index.

5. The method according to any one of claims 1 to 4, characterized in that: After the laser point cloud in the first point cloud set is processed by the advanced density clustering model and the second point cloud set is output, the method further includes: Acquire a plurality of point cloud data detected by the laser radar when scanning the surrounding environment in the previous frame; Get the 3D object detection model; The plurality of point cloud data detected by the current frame scanning and the plurality of point cloud data detected by the previous frame scanning are respectively subjected to smoke recognition by the 3D object detection model to obtain a first recognition result and a second recognition result, wherein the first recognition result and the second recognition result at least include one or more irregular polygonal objects; respectively calculating the area of ​​at least one irregular polygonal object in the first recognition result and the area of ​​at least one irregular polygonal object in the second recognition result; Comparing the area differences of the irregular polygonal objects in the first recognition result and the irregular polygonal objects in the second recognition result one by one, and calculating the intersection-and-union ratio of the areas of the irregular polygonal objects in the previous frame and the current frame; If the calculated intersection-and-union ratio is less than a threshold value, determining to use the laser point cloud on the area of ​​the irregular polygonal object corresponding to the intersection-and-union ratio as a third point cloud set; The laser point cloud in the third point cloud set and the laser point cloud in the second point cloud set are intersected to obtain at least one intersecting laser point cloud, and the at least one intersecting laser point cloud is determined to be a smoke point cloud.

6. The method according to claim 5, characterized in that Intersecting the laser point cloud in the third point cloud set with the laser point cloud in the second point cloud set to obtain at least one intersecting laser point cloud, and determining that the at least one intersecting laser point cloud is a smoke point cloud, including: Respectively obtaining intensity values ​​corresponding to each laser point cloud in the first recognition result and the second recognition result; A smoke point cloud index obtained by filtering according to the intensity value and density of the point cloud, wherein the smoke point cloud index is represented by i, i=0, 1, ...n; Obtain the corresponding smoke point cloud index according to the intersection-and-union ratio of adjacent frames in target recognition, where the corresponding smoke point cloud index is represented by j, where j=3, 4, ...m; Based on the smoke point cloud indexes i and j, an intersection operation is performed to obtain the index value of the target smoke point cloud. The index value of the target smoke point cloud is represented by k as follows: k=i∩j The smoke point cloud is determined according to the index value k of the target smoke point cloud.

7. The method according to any one of claims 1 to 4, characterized in that: Based on the comparison between the intensity value corresponding to each laser point cloud and the intensity threshold, the first point cloud set is screened and obtained, including: Determining whether the intensity value of each of the laser point clouds is less than the intensity threshold; If yes, all laser point clouds with intensity values ​​less than the intensity threshold are counted and index values ​​corresponding to the laser point clouds are recorded to generate the first point cloud set.

8. A smoke recognition device based on laser point cloud, characterized in that: The device comprises: A receiving module is used to receive a plurality of point cloud data detected by the laser radar when scanning the surrounding environment in the current frame, each point cloud data includes point cloud coordinates and intensity values ​​corresponding to the laser point cloud; A screening module, configured to screen and obtain a first point cloud set based on a comparison between an intensity value corresponding to each laser point cloud and an intensity threshold, wherein the intensity value of each laser point cloud in the first point cloud set is less than the intensity threshold; A processing module, used to obtain an initial density clustering model, configure hyperparameters in the initial density clustering model, and generate an advanced density clustering model, wherein the advanced density clustering model is used to identify and cluster point cloud data; The recognition module is used to process the laser point cloud in the first point cloud set through the advanced density clustering model, output a second point cloud set, and recognize that each point cloud data of multiple laser point clouds in the second point cloud set that are clustered into clusters is a smoke point cloud.

9. A computer device, characterized in that: comprising a memory and a processor, wherein the memory and the processor are connected; The memory stores computer instructions, and the processor executes the smoke recognition method of laser point cloud according to any one of claims 1 to 7 by executing the computer instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the laser point cloud smoke recognition method according to any one of claims 1 to 7.

Images