CN115271096B - Point cloud processing and machine learning model training method and device and automatic driving vehicle - Google Patents
Point cloud processing and machine learning model training method and device and automatic driving vehicle Download PDFInfo
- Publication number
- CN115271096B CN115271096B CN202210894794.5A CN202210894794A CN115271096B CN 115271096 B CN115271096 B CN 115271096B CN 202210894794 A CN202210894794 A CN 202210894794A CN 115271096 B CN115271096 B CN 115271096B
- Authority
- CN
- China
- Prior art keywords
- point cloud
- target point
- machine learning
- target
- learning model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000012549 training Methods 0.000 title claims abstract description 64
- 238000010801 machine learning Methods 0.000 title claims abstract description 60
- 238000000034 method Methods 0.000 title claims abstract description 57
- 238000012545 processing Methods 0.000 title claims description 33
- 238000003672 processing method Methods 0.000 claims abstract description 21
- 239000003595 mist Substances 0.000 claims description 24
- 239000000428 dust Substances 0.000 claims description 20
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims description 20
- 239000011159 matrix material Substances 0.000 claims description 10
- 230000004044 response Effects 0.000 claims description 8
- 230000002452 interceptive effect Effects 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 6
- 238000000354 decomposition reaction Methods 0.000 claims description 5
- 238000013473 artificial intelligence Methods 0.000 abstract description 3
- 238000003066 decision tree Methods 0.000 description 23
- 238000004422 calculation algorithm Methods 0.000 description 14
- 238000010586 diagram Methods 0.000 description 13
- 230000008569 process Effects 0.000 description 13
- 238000004891 communication Methods 0.000 description 10
- 238000004590 computer program Methods 0.000 description 10
- 238000001514 detection method Methods 0.000 description 9
- 238000007637 random forest analysis Methods 0.000 description 8
- 239000013598 vector Substances 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 230000003287 optical effect Effects 0.000 description 3
- 230000003044 adaptive effect Effects 0.000 description 2
- 230000002776 aggregation Effects 0.000 description 2
- 238000004220 aggregation Methods 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 206010039203 Road traffic accident Diseases 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 238000013475 authorization Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000001953 sensory effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Image Analysis (AREA)
Abstract
The disclosure provides a point cloud processing method, relates to the technical field of artificial intelligence, and particularly relates to the technical field of automatic driving and intelligent traffic. The specific implementation scheme is as follows: acquiring respective characteristics of a plurality of target point clouds; classifying each target point cloud by utilizing a machine learning model according to the characteristics of each of the plurality of target point clouds to obtain a category of each target point cloud, wherein the category is used for indicating whether the target point cloud is noise or not; and rejecting the target point cloud classified as noise from the plurality of target point clouds. The disclosure also provides a training method and device of the machine learning model, electronic equipment, storage medium and automatic driving vehicle.
Description
Technical Field
The present disclosure relates to the field of artificial intelligence, and more particularly to automated driving and intelligent transportation technologies. More specifically, the disclosure provides a point cloud processing method, a training method of a machine learning model, an apparatus, an electronic device, a storage medium and an automatic driving vehicle.
Background
In the field of automatic driving, sensing the surrounding environment by using a laser radar is an important means for ensuring the normal running of an automatic driving vehicle. For example, the lidar may collect point cloud data of the surrounding environment, and sense the surrounding environment by detecting point clouds (e.g., road point clouds, pedestrian point clouds, etc.) of respective objects from the point cloud data of the surrounding environment.
Disclosure of Invention
The disclosure provides a point cloud processing method, a training device, training equipment, training storage media and an automatic driving vehicle.
According to a first aspect, there is provided a point cloud processing method, the method comprising: acquiring respective characteristics of a plurality of target point clouds; classifying each target point cloud by utilizing a machine learning model according to the characteristics of each of the plurality of target point clouds to obtain a category of each target point cloud, wherein the category is used for indicating whether the target point cloud is noise or not; and rejecting the target point cloud classified as noise from the plurality of target point clouds.
According to a second aspect, there is provided a method of training a machine learning model, the method comprising: obtaining M sample point clouds, wherein each sample point cloud comprises N dimension characteristics and a class label for indicating whether the sample is noise, M is a positive integer, and N is a positive integer; responding to a kth training request, selecting M sample point clouds from M sample point clouds, and training by using the characteristics of N dimensions of the M sample point clouds to obtain a kth classifier, wherein M is a positive integer smaller than M, and N is a positive integer smaller than N; and determining K classifiers as machine learning models in response to the K training completions, wherein K is a positive integer, and K is greater than or equal to 1 and less than or equal to K.
According to a third aspect, there is provided a point cloud processing apparatus, the apparatus comprising: the first acquisition module is used for acquiring the characteristics of each of the target point clouds; the classification module is used for classifying each target point cloud by utilizing a machine learning model according to the characteristics of each of the target point clouds to obtain the category of each target point cloud, wherein the category is used for indicating whether the target point cloud is noise or not; and the processing module is used for eliminating the target point cloud classified as noise from the plurality of target point clouds.
According to a fourth aspect, there is provided a training apparatus of a machine learning model, the apparatus comprising: the second acquisition module is used for acquiring M sample point clouds, wherein each sample point cloud comprises N dimension characteristics and a class label for indicating whether the sample is noise, M is a positive integer, and N is a positive integer; the training module is used for responding to a kth training request, selecting M sample point clouds from M sample point clouds, and training by using the characteristics of N dimensions of the M sample point clouds to obtain a kth classifier, wherein M is a positive integer smaller than M, and N is a positive integer smaller than N; and a determining module for determining K classifiers as machine learning models in response to completion of K times of training, wherein K is a positive integer, and K is 1 or more and K or less.
According to a fifth aspect, there is provided an electronic device comprising: at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform a method provided in accordance with the present disclosure.
According to a sixth aspect, there is provided a non-transitory computer readable storage medium storing computer instructions for causing a computer to perform a method provided according to the present disclosure.
According to a seventh aspect, there is provided a computer program product comprising a computer program which, when executed by a processor, implements a method provided according to the present disclosure.
According to an eighth aspect, there is provided an autonomous vehicle comprising an electronic device provided according to the fifth aspect.
It should be understood that the description in this section is not intended to identify key or critical features of the embodiments of the disclosure, nor is it intended to be used to limit the scope of the disclosure. Other features of the present disclosure will become apparent from the following specification.
Drawings
The drawings are for a better understanding of the present solution and are not to be construed as limiting the present disclosure. Wherein:
FIG. 1 is a schematic diagram of an exemplary system architecture to which a point cloud processing method and a training method of a machine learning model may be applied, according to one embodiment of the present disclosure;
FIG. 2 is a flow chart of a point cloud processing method according to one embodiment of the present disclosure;
FIG. 3 is a flow chart of a point cloud processing method according to another embodiment of the present disclosure;
FIG. 4 is a flow chart of a method of training a machine learning model according to one embodiment of the present disclosure;
FIG. 5A is a schematic diagram of a decision tree according to one embodiment of the present disclosure;
FIG. 5B is a flow chart of a method of generating a decision tree according to one embodiment of the present disclosure;
FIG. 6 is a block diagram of a point cloud processing device according to one embodiment of the present disclosure;
FIG. 7 is a block diagram of a training apparatus of a machine learning model according to one embodiment of the present disclosure;
FIG. 8 is a block diagram of an electronic device of a point cloud processing method and/or a training method of a machine learning model according to one embodiment of the present disclosure;
fig. 9 is a schematic diagram of an autonomous vehicle according to one embodiment of the present disclosure.
Detailed Description
Exemplary embodiments of the present disclosure are described below in conjunction with the accompanying drawings, which include various details of the embodiments of the present disclosure to facilitate understanding, and should be considered as merely exemplary. Accordingly, one of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.
The laser radar has the functions of detecting the distance to the surrounding environment and positioning the position of an object, and mainly measures the distance by emitting a laser sensing beam and according to the time when the beam returns to a sensor after contacting the object. Because the water mist, dust and the like in the air have great influence on the propagation of light beams, the laser radar is sensitive to the water mist and the dust in the air, and the collected point cloud data can cause false detection of objects due to the fact that the collected point cloud data contain water mist data or dust data. Such as false detection of water mist or dust as a real object (e.g., a vehicle, a pedestrian, etc.), or false detection of a real object as water mist or dust. The false sense of the object may cause the vehicle to be braked suddenly, and thus may cause serious traffic accidents, so that the identification of water mist or dust from the point cloud data is critical for safe driving.
A method for identifying water mist from point cloud data is to distinguish real objects from water mist by utilizing geometrical distribution characteristics of point cloud. The more uniform the distribution of the point cloud between the points, the greater the probability of being water mist, the less the probability of being water mist, the point cloud in which the distribution of the points is concentrated on one plane. However, since the distribution characteristics of the point clouds located at different distances are different (for example, the point clouds near are dense and the point clouds far are sparse), the error rate of identifying the water mist by using the method is high.
In the technical scheme of the disclosure, the related processes of collecting, storing, using, processing, transmitting, providing, disclosing and the like of the personal information of the user accord with the regulations of related laws and regulations, and the public order colloquial is not violated.
In the technical scheme of the disclosure, the authorization or consent of the user is obtained before the personal information of the user is obtained or acquired.
FIG. 1 is a schematic diagram of an exemplary system architecture to which a point cloud processing method and a training method of a machine learning model may be applied, according to one embodiment of the present disclosure. It should be noted that fig. 1 is only an example of a system architecture to which embodiments of the present disclosure may be applied to assist those skilled in the art in understanding the technical content of the present disclosure, but does not mean that embodiments of the present disclosure may not be used in other devices, systems, environments, or scenarios.
As shown in fig. 1, a system architecture 100 according to this embodiment may include terminal devices 101, 102, 103, a network 104, and a server 105. The network 104 is used as a medium to provide communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired and/or wireless communication links, and the like.
The user may interact with the server 105 via the network 104 using the terminal devices 101, 102, 103 to receive or send messages or the like. The terminal devices 101, 102, 103 may be a variety of electronic devices including, but not limited to, smartphones, tablets, laptop portable computers, and the like.
At least one of the point cloud processing method and the training method of the machine learning model provided by the embodiments of the present disclosure may be generally performed by the server 105. Accordingly, the point cloud processing apparatus and the training apparatus of the machine learning model provided in the embodiments of the present disclosure may be generally disposed in the server 105. The point cloud processing method and the training method of the machine learning model provided by the embodiments of the present disclosure may also be performed by a server or a server cluster that is different from the server 105 and is capable of communicating with the terminal devices 101, 102, 103 and/or the server 105. Accordingly, the point cloud processing apparatus and the training apparatus of the machine learning model provided by the embodiments of the present disclosure may also be provided in a server or a server cluster that is different from the server 105 and is capable of communicating with the terminal devices 101, 102, 103 and/or the server 105.
Fig. 2 is a flow chart of a point cloud processing method according to one embodiment of the present disclosure.
As shown in fig. 2, the point cloud processing method 200 may include operations S210 to S230.
In operation S210, characteristics of each of a plurality of target point clouds are acquired.
The plurality of target point clouds may be detected from point cloud data acquired by a lidar. For example, the point cloud data collected by the laser radar comprises a point cloud set of all objects in an operating environment of the automatic driving vehicle, the point cloud of each object is detected from the point cloud set, the point cloud of each object is divided, and the identification of each object in the environment is realized, and the processing process can be called target detection.
The plurality of target point clouds of target detection may include pedestrian point clouds, vehicle point clouds, tree point clouds, road point clouds, water mist point clouds, dust point clouds, and the like. The point clouds of pedestrians, vehicles, trees or other obstacles play a role in deciding the motion of the automatic driving vehicle, and the point clouds of the objects can be called as the point clouds of real objects. While the cloud of water mist or dust should not affect the decision to automatically drive the vehicle, these objects may act as interfering objects that affect the automatic driving, the cloud of interfering objects may be referred to as a noisy cloud.
The characteristics of each target point cloud may include the position (three-dimensional coordinates), center point (coordinate position), roughness, etc. of each point in the target point cloud. The center point position may be an average of three-dimensional coordinates of points in the target point cloud. Roughness can be characterized by the divergent index, the planar index of the cloud of target points.
For example, three-dimensional coordinates of each point in the target point cloud may form a location feature matrix, singular value decomposition is performed on the location feature matrix, a plurality of singular feature values of the target point cloud may be obtained, and a divergent index and a planar index of the target point cloud may be calculated according to the plurality of singular feature values.
For example, the divergence index of the target point cloud may be expressed by the following formula (one).
The planeness index of the target point cloud can be expressed by the following formula (two).
Wherein S is λ Represents the divergence form index, P λ Represents the planeness index lambda 1 、λ 2 And lambda (lambda) 3 Is three singular eigenvalues of the target point cloud, and lambda 1 、λ 2 And lambda (lambda) 3 Sequentially decreasing in size.
In operation S220, according to the characteristics of each of the multiple target point clouds, classifying each target point cloud by using a machine learning model, to obtain a class of each target point cloud.
For example, the machine learning model may be trained based on random forest (also known as random decision forest) algorithms. The random forest algorithm may be a Boosting algorithm (e.g., adaptive Boosting, adaptive enhancement algorithm) or a Bagging algorithm (Bootstrap Aggregation, bootstrap aggregation algorithm, simply Bagging algorithm).
The machine learning model based on the random forest algorithm may include a plurality of decision trees, each acting as a classifier. And inputting the characteristics of the target point cloud into a machine learning model, and classifying the characteristics of the target point cloud by a plurality of decision trees to obtain a plurality of classification results. The classification result may include a class of the target point cloud, which may be used to indicate whether the target point cloud is noise. For example, the category is "1" indicating that the target point cloud is noise (water mist or dust), the category is "0" indicating that the target point cloud is not noise (the target point cloud is a real object).
For example, the class of the target point cloud may be determined from classification results of multiple decision trees. For example, the classification result with the highest number of the same class is selected as the class of the target point cloud, or the weighted average of all classification results is used as the class of the target point cloud.
In operation S230, the target point cloud classified as noise is removed from the plurality of target point clouds.
For example, aiming at a target point cloud with noise, the target point cloud can be determined to be data generated after an interference object such as water mist or dust is scanned, and the water mist point cloud or the dust point cloud is removed from a plurality of target point clouds, so that danger caused by emergency braking of an automatic driving vehicle can be avoided, and safe driving of the automatic driving vehicle is ensured.
According to the embodiment of the disclosure, the machine learning model is used for classifying the target point clouds, and the target point clouds with noise class are identified from the plurality of target point clouds, so that the accuracy rate of the point cloud identification can be improved compared with the method for identifying noise based on the geometric distribution characteristics.
It can be appreciated that in rainy days or in flying dust weather, the embodiment of the disclosure can filter water mist or dust in the air from the point cloud detection result, so that the safety of automatic driving is effectively improved.
Fig. 3 is a flow chart of a point cloud processing method according to another embodiment of the present disclosure.
As shown in fig. 3, the point cloud processing method may include operations S310 to S350.
In operation S310, a plurality of target point clouds are acquired.
For example, the plurality of target point clouds may be obtained by performing target detection on point cloud data acquired by the lidar. The plurality of target point clouds may include pedestrian point clouds, vehicle point clouds, tree point clouds, road point clouds, water mist point clouds, dust point clouds, and the like.
In operation S320, the target point cloud is preprocessed.
For example, for each target point cloud, the features of each point in the target point cloud are preprocessed, which may include normalization processing. The characteristics of each point may include coordinates, color, reflectance values, etc., and at least one of the coordinates, color, reflectance values of each point may be normalized.
It can be understood that the normalization processing is performed on the target point cloud, so that the subsequent calculation of the point cloud features is simpler and more convenient, the calculated features have smaller disturbance, and the calculation of machine learning is facilitated.
In operation S330, feature vectors of the target point cloud are calculated.
For example, the characteristics of the target point cloud may include three-dimensional coordinates, center point coordinates, roughness, etc. of points in the target point cloud. The center point coordinates may be an average of three-dimensional coordinates of points in the target point cloud. The roughness includes a divergent index, a planar index.
The three-dimensional coordinates, the center point coordinates and the roughness characteristics of each point in the target point cloud can be vectorized, and the characteristic vector of the target point cloud is obtained.
In operation S340, the target point cloud is classified by using the machine learning model, and the class of the target point cloud is obtained.
For example, feature vectors of the target point cloud are input to a machine learning model, resulting in a class of the target point cloud that indicates whether the target point cloud is noise. The specific implementation of operation S340 may refer to operation S220, and will not be described herein.
In operation S350, the target point cloud classified as noise is removed from the plurality of target point clouds.
The specific implementation of operation S350 is similar to that of operation S230, and will not be described again.
According to the embodiment of the disclosure, the target point cloud is preprocessed, the feature vector is calculated for the preprocessed target point cloud, and the class of the target point cloud is determined based on the feature vector by using the machine learning model, so that the real object and noise in the point cloud data can be effectively distinguished.
Fig. 4 is a flow chart of a method of training a machine learning model according to one embodiment of the present disclosure.
As shown in fig. 4, the training method 400 of the machine learning model includes operations S410 to S430.
In operation S410, M sample point clouds are acquired.
Where M is a positive integer, e.g., m=10ten thousand. 10 ten thousand sample point clouds are acquired, each sample point cloud including features of N (N is a positive integer, e.g., n=5) dimensions and a class label indicating whether the sample is noise. The characteristics of the N dimensions may include coordinates, center point coordinates, divergent indices, and planar indices of points in the sample point cloud.
In operation S420, in response to the kth training request, M sample point clouds are selected from the M sample point clouds, and training is performed using the n-dimensional features of the M sample point clouds, so as to obtain the kth classifier.
In operation S430, K classifiers are determined as machine learning models in response to the completion of the K times of training.
For example, the M sample point clouds described above may be used to train based on a random decision forest algorithm (e.g., bagging algorithm) resulting in a machine learning model that contains multiple decision trees (each as a classifier).
For example, each training takes M (e.g., m=50) samples from M (e.g., m=10 ten thousand) sample point clouds, and uses the features of n (e.g., n=3) dimensions of the M samples for training the classifier, resulting in a trained classifier. Through K (for example, K=100) round training, K trained classifiers are obtained, and the K trained classifiers form a machine learning model.
The kth training (K is greater than or equal to 1 and less than or equal to K, that is, each time) selects M (for example, m=50) sample point clouds from M (for example, m=10 ten thousand) sample point clouds, and the used N (for example, n=3) dimension features are also selected randomly from N (for example, n=5) dimension features.
According to the embodiment of the disclosure, the sample point cloud is used for training based on a random forest algorithm, so that a machine learning model is obtained, and real objects and noise in point cloud data can be classified by using the machine learning model.
Each decision tree of the machine learning model serves as a classifier, and each training process corresponds to the tree generation process. The training method of the machine learning model provided by the present disclosure is described in detail below with reference to fig. 5A to 5B.
Fig. 5A is a schematic diagram of a decision tree according to one embodiment of the present disclosure.
As shown in FIG. 5A, in an example, decision tree 501 includes a plurality of nodes, such as nodes 1-5. Node 1 may be a root node, nodes 2-3 are children of node 1, and nodes 2-3 may be referred to as primary children. Nodes 4-5 are children of node 2, and nodes 4-5 may be referred to as secondary children.
For example, decision tree 501 is the result of training of the current round. In the current training process, each node correspondingly processes the characteristics of one dimension, the tree generation process is equivalent to the process of distributing the characteristics of multiple dimensions to the corresponding nodes for processing, and the specific mode of the characteristic distribution is not unique. For example, the characteristics of the sample point cloud in multiple dimensions include a characteristic a, a characteristic B, a characteristic C, a characteristic D, and a characteristic E, and there are various allocation manners for allocating the characteristics a to E to the nodes 1 to 5. The decision tree 502 shown in fig. 5A illustrates an allocation manner in which the node 1 corresponds to the feature a, the node 2 corresponds to the feature C, the node 3 corresponds to the feature B, the node 4 corresponds to the feature D, and the node 5 corresponds to the feature E. Similarly, decision tree 503 illustrates another way of allocation. Other allocation methods are also possible, and this embodiment is not listed one by one. The optimal allocation manner may be selected from a plurality of allocation manners as the tree obtained by the current training, for example, an allocation manner that makes the prediction accuracy of the tree highest (for example, the prediction accuracy of the decision tree 503 is highest) is selected as the tree obtained by the current training.
The process of generating the decision tree will be described below using decision tree 502 as an example.
For example, for decision tree 502, each node is generated based on a random forest algorithm. The method comprises the steps that a characteristic A of a sample point cloud is distributed to a node 1, the node 1 calculates the characteristic A based on a random forest algorithm, and partial characteristics (such as a characteristic C and a characteristic B) are selected from the residual characteristics according to a preset criterion (such as calculating information gain according to a calculation result and a sample label and according to the maximum information gain) and distributed to next-level sub-nodes (node 2 and node 3) respectively. And respectively calculating the feature B and the feature C by the node 2 and the node 3, selecting part of features from the rest features according to the calculation result and the preset criteria to be distributed to the next level of sub-nodes, and sequentially distributing the features of multiple dimensions by analogy until all the features of the dimensions are distributed, the depth of the tree reaches an upper limit (for example, 3), or the prediction effect of the tree reaches a threshold (for example, the accuracy is more than 50%).
The above process of selecting a part of the features from the remaining features to be allocated to the next level of child nodes may be referred to as a feature splitting process, which is also a child node generating process.
Fig. 5B is a flow chart of a method of generating a decision tree according to one embodiment of the present disclosure.
As shown in fig. 5B, the training method of the machine learning model includes operations S510 to S530.
In operation S510, a feature to be split is selected.
For example, operation S510 may include operations S511 to S512.
In operation S511, the current plurality of features is traversed, and each feature is determined as an optimal splitting manner of the feature to be split.
In operation S512, the feature to be split is determined according to the splitting effect of each feature as the feature to be split.
For example, the plurality of features refers to features of multiple dimensions, which in one example may include feature a, feature B, feature C, feature D, and feature E. In the case of a first split of the decision tree, the current plurality of features includes all of the features described above, and in the case of a non-first split, the current plurality of features refers to features that can be used to perform feature splitting (e.g., in the case of a second split, the current features include feature C and feature B).
The current plurality of features may be traversed to determine each feature as an optimal allocation (e.g., the highest predictive accuracy of the tree) for the feature to be split. And determining the feature to be split according to the distribution mode corresponding to the tree with the best splitting effect (the tree with the highest prediction accuracy) in the optimal distribution mode corresponding to each feature.
For example, in the case of a first splitting of a decision tree, a feature to be split (e.g., feature a) is selected from all features to be allocated to the root node via operations S511-S512. In the case of the decision tree splitting for the second time, the feature to be split (for example, feature C) is selected from the features (for example, feature C and feature B) corresponding to the current node to be allocated to the first level child node through operations S511 to S512.
In operation S520, feature splitting is performed to generate child nodes.
For example, a partial feature is selected from the features of the remaining dimensions to be assigned to the next level child node. For example, in the case of a first splitting of the decision tree, part of the features (e.g., C and B) are selected from the remaining dimensional features (feature B, feature C, feature D, feature E) and assigned to the level one child node. In the case of a second splitting of the decision tree, partial features (e.g., feature D and feature E) are selected from the features of the remaining dimensions (e.g., feature D and feature E) to be assigned to the secondary child nodes.
It can be understood that in the case that the decision tree is a binary tree, the number of partial features selected for each feature splitting includes two, and correspondingly, the number of generated child nodes is also two.
In operation S530, it is determined whether splitting is continued, if yes, operation S510 is performed back, otherwise node splitting is ended.
For example, the condition for determining whether to continue splitting may include at least one of the feature allocation being complete for all dimensions, or the depth of the current tree reaching an upper limit, or the predictive effect of the current tree reaching a threshold (e.g., accuracy greater than 50%).
According to an embodiment of the disclosure, the disclosure further provides a point cloud processing device and a training device of the machine learning model.
Fig. 6 is a block diagram of a point cloud processing device according to one embodiment of the present disclosure.
As shown in fig. 6, the point cloud processing apparatus 600 includes a first acquisition module 601, a classification module 602, and a processing module 603.
The first obtaining module 601 is configured to obtain characteristics of each of the multiple target point clouds.
The classification module 602 is configured to classify each target point cloud by using a machine learning model according to respective characteristics of the plurality of target point clouds, to obtain a class of each target point cloud, where the class is used to indicate whether the target point cloud is noise.
The processing module 603 is configured to reject the target point cloud classified as noise from the plurality of target point clouds.
For example, a machine learning model includes a plurality of classifiers; the classification module comprises a classification unit and a first determination unit.
The classification unit is used for classifying the characteristics of each target point cloud by utilizing a plurality of classifiers to obtain a plurality of classification results.
The first determining unit is used for determining the category of each target point cloud according to a plurality of classification results.
The first acquisition module comprises a processing unit and a second determination unit.
The processing unit is used for carrying out normalization processing on the characteristics of each of a plurality of points in the target point cloud aiming at each target point cloud.
The second determining unit is used for determining characteristics of the target point cloud according to the normalized multiple points for each target point cloud.
For example, the noise includes a point cloud of the interfering object including at least one of mist and dust.
For example, the characteristics of the target point cloud include: at least one of a location of each point in the target point cloud, a center point location, a divergent index, and a planar index.
Fig. 7 is a block diagram of a training apparatus of a machine learning model according to one embodiment of the present disclosure.
As shown in fig. 7, the training 700 of the machine learning model may include a second acquisition module 701, a training module 702, and a determination module 703.
The second acquisition module 701 is configured to acquire M sample point clouds. Each sample point cloud includes N-dimensional features and a class label indicating whether the sample is noise, where M is a positive integer and N is a positive integer.
The training module 702 is configured to select M sample point clouds from M sample point clouds in response to a kth training request, and perform training using N-dimensional features of the M sample point clouds to obtain a kth classifier, where M is a positive integer less than M, and N is a positive integer less than N.
The determining module 703 is configured to determine K classifiers as machine learning models in response to completion of K times of training, where K is a positive integer, and K is greater than or equal to 1 and less than or equal to K.
The noise includes a point cloud of the interfering object including at least one of mist and dust.
The N-dimensional features include at least one of a location of each point in the sample point cloud, a center point location, a divergent index, and a planar index.
According to embodiments of the present disclosure, the present disclosure also provides an electronic device, an autonomous vehicle, a readable storage medium and a computer program product.
Fig. 8 illustrates a schematic block diagram of an example electronic device 800 that may be used to implement embodiments of the present disclosure. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular telephones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be exemplary only, and are not meant to limit implementations of the disclosure described and/or claimed herein.
As shown in fig. 8, the apparatus 800 includes a computing unit 801 that can perform various appropriate actions and processes according to a computer program stored in a Read Only Memory (ROM) 802 or a computer program loaded from a storage unit 808 into a Random Access Memory (RAM) 803. In the RAM 803, various programs and data required for the operation of the device 800 can also be stored. The computing unit 801, the ROM 802, and the RAM 803 are connected to each other by a bus 804. An input/output (I/O) interface 805 is also connected to the bus 804.
Various components in device 800 are connected to I/O interface 805, including: an input unit 806 such as a keyboard, mouse, etc.; an output unit 807 such as various types of displays, speakers, and the like; a storage unit 808, such as a magnetic disk, optical disk, etc.; and a communication unit 809, such as a network card, modem, wireless communication transceiver, or the like. The communication unit 809 allows the device 800 to exchange information/data with other devices via a computer network such as the internet and/or various telecommunication networks.
The computing unit 801 may be a variety of general and/or special purpose processing components having processing and computing capabilities. Some examples of computing unit 801 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various specialized Artificial Intelligence (AI) computing chips, various computing units running machine learning model algorithms, a Digital Signal Processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 801 performs the various methods and processes described above, such as a point cloud processing method and/or a training method of a machine learning model. For example, in some embodiments, the point cloud processing method and/or the training method of the machine learning model may be implemented as a computer software program tangibly embodied on a machine-readable medium, such as the storage unit 808. In some embodiments, part or all of the computer program may be loaded and/or installed onto device 800 via ROM 802 and/or communication unit 809. When the computer program is loaded into the RAM 803 and executed by the computing unit 801, one or more steps of the point cloud processing method and/or the training method of the machine learning model described above may be performed. Alternatively, in other embodiments, the computing unit 801 may be configured to perform the point cloud processing method and/or the training method of the machine learning model in any other suitable manner (e.g., by means of firmware).
FIG. 9 illustrates a schematic block diagram of an example autonomous vehicle 900 that may be used to implement embodiments of the present disclosure.
As shown in fig. 9, the autonomous vehicle 900 includes a point cloud collecting device 901 and an electronic device 902. The point cloud acquisition device 901 and the electronic device 902 may communicate via a wired or wireless connection.
The point cloud acquisition device 901 includes, for example, a LiDAR (Light Detection and Ranging, laser detection and ranging) sensor, which can continuously emit a laser sensor beam during the driving of the autonomous vehicle 900 to scan for a point cloud of surrounding objects. The point cloud acquisition device 901 may send the acquired point cloud data to the electronic device 902, where the electronic device 902 implements, for example, the point cloud processing method and/or the training method of the machine learning model described above, to obtain the target point cloud after noise is removed. The automatic driving vehicle 900 determines a motion track according to the target point cloud after noise is removed, so that phenomena such as sudden braking and the like caused by noise influence can be avoided, and the safety of automatic driving is ensured.
It can be appreciated that the electronic device 902 is disposed in the autopilot vehicle 900, which can save data transmission time, improve data processing efficiency, and further guarantee autopilot safety compared to an external server.
Various implementations of the systems and techniques described here above may be implemented in digital electronic circuitry, integrated circuit systems, field Programmable Gate Arrays (FPGAs), application Specific Integrated Circuits (ASICs), application Specific Standard Products (ASSPs), systems On Chip (SOCs), complex Programmable Logic Devices (CPLDs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs, the one or more computer programs may be executed and/or interpreted on a programmable system including at least one programmable processor, which may be a special purpose or general-purpose programmable processor, that may receive data and instructions from, and transmit data and instructions to, a storage system, at least one input device, and at least one output device.
Program code for carrying out methods of the present disclosure may be written in any combination of one or more programming languages. These program code may be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus such that the program code, when executed by the processor or controller, causes the functions/operations specified in the flowchart and/or block diagram to be implemented. The program code may execute entirely on the machine, partly on the machine, as a stand-alone software package, partly on the machine and partly on a remote machine or entirely on the remote machine or server.
In the context of this disclosure, a machine-readable medium may be a tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. The machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to a user; and a keyboard and pointing device (e.g., a mouse or trackball) by which a user can provide input to the computer. Other kinds of devices may also be used to provide for interaction with a user; for example, feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form, including acoustic input, speech input, or tactile input.
The systems and techniques described here can be implemented in a computing system that includes a background component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such background, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), wide Area Networks (WANs), and the internet.
The computer system may include a client and a server. The client and server are typically remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other.
It should be appreciated that various forms of the flows shown above may be used to reorder, add, or delete steps. For example, the steps recited in the present disclosure may be performed in parallel, sequentially, or in a different order, provided that the desired results of the disclosed aspects are achieved, and are not limited herein.
The above detailed description should not be taken as limiting the scope of the present disclosure. It will be apparent to those skilled in the art that various modifications, combinations, sub-combinations and alternatives are possible, depending on design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present disclosure are intended to be included within the scope of the present disclosure.
Claims (11)
1. A point cloud processing method, comprising:
acquiring respective characteristics of a plurality of target point clouds;
classifying each target point cloud by using a machine learning model according to the characteristics of each of the plurality of target point clouds to obtain a category of each target point cloud, wherein the category is used for indicating whether the target point cloud is noise or not; and
removing target point clouds classified as noise from the plurality of target point clouds;
wherein the noise comprises an interfering object comprising at least one of mist and dust; the characteristics of the target point cloud include: the position, the central point position, the divergent index and the planar index of each point in the target point cloud;
the acquiring the characteristics of each of the plurality of target point clouds comprises: for each of the cloud of target points,
forming a position feature matrix according to the positions of each point in the target point cloud;
singular value decomposition is carried out on the position feature matrix to obtain a plurality of singular value features of the target point cloud;
and calculating a divergent index and a planar index of the target point cloud according to the singular value features.
2. The method of claim 1, wherein the machine learning model comprises a plurality of classifiers; classifying each target point cloud by using a machine learning model according to the characteristics of each of the plurality of target point clouds, and obtaining the category of each target point cloud comprises: for each of the cloud of target points,
classifying the characteristics of the target point cloud by utilizing the plurality of classifiers to obtain a plurality of classification results; and
and determining the category of the target point cloud according to the classification results.
3. The method of claim 1 or 2, wherein the acquiring characteristics of each of a plurality of target point clouds comprises: for each of the cloud of target points,
normalizing the characteristics of each of a plurality of points in the target point cloud; and
and determining the characteristics of the target point cloud according to the plurality of points after normalization processing.
4. A method of training a machine learning model, comprising:
obtaining M sample point clouds, wherein each sample point cloud comprises N dimension characteristics and a class label for indicating whether the sample is noise, M is a positive integer, and N is a positive integer;
responding to a kth training request, selecting M sample point clouds from M sample point clouds, and training by using the characteristics of N dimensions of the M sample point clouds to obtain a kth classifier, wherein M is a positive integer smaller than M, and N is a positive integer smaller than N; and
in response to completion of K training, determining K classifiers as the machine learning model, wherein K is a positive integer, and K is greater than or equal to 1 and less than or equal to K;
the noise comprises an interference object, the interference object comprises at least one of water mist and dust, and the N dimensional characteristics comprise positions of points in a sample point cloud, a center point position, a divergent index and a planar index;
the method further comprises the steps of: for each of the sample point clouds,
forming a position feature matrix according to the positions of each point in the sample point cloud;
performing singular value decomposition on the position feature matrix to obtain a plurality of singular value features of the sample point cloud;
and calculating the divergent index and the planar index of the sample point cloud according to the singular value features.
5. A point cloud processing apparatus, comprising:
the first acquisition module is used for acquiring the characteristics of each of the target point clouds;
the classification module is used for classifying each target point cloud by utilizing a machine learning model according to the characteristics of each target point cloud to obtain the category of each target point cloud, wherein the category is used for indicating whether the target point cloud is noise or not; and
the processing module is used for removing the target point cloud classified as noise from the plurality of target point clouds;
wherein the noise comprises an interfering object comprising at least one of mist and dust; the characteristics of the target point cloud include: the position, the central point position, the divergent index and the planar index of each point in the target point cloud;
the first acquisition module is used for forming a position feature matrix according to the positions of each point in each target point cloud; singular value decomposition is carried out on the position feature matrix to obtain a plurality of singular value features of the target point cloud; and calculating a divergent index and a planar index of the target point cloud according to the singular value features.
6. The apparatus of claim 5, wherein the machine learning model comprises a plurality of classifiers; the classification module comprises:
the classification unit is used for classifying the characteristics of each target point cloud by utilizing the plurality of classifiers to obtain a plurality of classification results; and
and the first determining unit is used for determining the category of each target point cloud according to the plurality of classification results.
7. The apparatus of claim 5 or 6, wherein the first acquisition module comprises:
the processing unit is used for carrying out normalization processing on the characteristics of each of a plurality of points in each target point cloud; and
and the second determining unit is used for determining the characteristics of each target point cloud according to the plurality of points after normalization processing.
8. A training apparatus for a machine learning model, comprising:
the second acquisition module is used for acquiring M sample point clouds, wherein each sample point cloud comprises N dimension characteristics and a class label for indicating whether the sample is noise, M is a positive integer, and N is a positive integer;
the training module is used for responding to a kth training request, selecting M sample point clouds from M sample point clouds, and training by using the characteristics of N dimensions of the M sample point clouds to obtain a kth classifier, wherein M is a positive integer smaller than M, and N is a positive integer smaller than N; and
the determining module is used for determining K classifiers as the machine learning model in response to the completion of K times of training, wherein K is a positive integer, and K is greater than or equal to 1 and less than or equal to K;
the noise comprises an interference object, the interference object comprises at least one of water mist and dust, and the N dimensional characteristics comprise positions of points in a sample point cloud, a center point position, a divergent index and a planar index;
the apparatus further comprises:
the third acquisition module is used for forming a position feature matrix according to the positions of each point in each sample point cloud; performing singular value decomposition on the position feature matrix to obtain a plurality of singular value features of the sample point cloud; and calculating the divergent index and the planar index of the sample point cloud according to the singular value features.
9. An electronic device, comprising:
at least one processor; and
a memory communicatively coupled to the at least one processor; wherein,,
the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 4.
10. A non-transitory computer readable storage medium storing computer instructions for causing the computer to perform the method of any one of claims 1 to 4.
11. An autonomous vehicle comprising the electronic device of claim 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210894794.5A CN115271096B (en) | 2022-07-27 | 2022-07-27 | Point cloud processing and machine learning model training method and device and automatic driving vehicle |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210894794.5A CN115271096B (en) | 2022-07-27 | 2022-07-27 | Point cloud processing and machine learning model training method and device and automatic driving vehicle |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115271096A CN115271096A (en) | 2022-11-01 |
| CN115271096B true CN115271096B (en) | 2023-08-18 |
Family
ID=83771446
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210894794.5A Active CN115271096B (en) | 2022-07-27 | 2022-07-27 | Point cloud processing and machine learning model training method and device and automatic driving vehicle |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115271096B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116311172B (en) * | 2023-05-17 | 2023-09-22 | 九识(苏州)智能科技有限公司 | Training method, device, equipment and storage medium of 3D target detection model |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109063789A (en) * | 2018-09-07 | 2018-12-21 | 北京数字绿土科技有限公司 | The classification method of shaft tower and power line, device and system |
| WO2021008202A1 (en) * | 2019-07-16 | 2021-01-21 | 浙江大学 | Method for kernel support vector machine target classification based on millimeter-wave radar point cloud features |
| CN112513679A (en) * | 2020-04-15 | 2021-03-16 | 华为技术有限公司 | Target identification method and device |
| CN112837239A (en) * | 2021-02-01 | 2021-05-25 | 意诺科技有限公司 | Residual image eliminating method and device for millimeter wave radar and computer readable medium |
| CN113516052A (en) * | 2021-05-21 | 2021-10-19 | 同济大学 | Imaging millimeter wave radar point cloud target classification method based on machine learning |
| CN113763263A (en) * | 2021-07-27 | 2021-12-07 | 华能伊敏煤电有限责任公司 | Water mist tail gas noise treatment method based on point cloud tail gas filtering technology |
-
2022
- 2022-07-27 CN CN202210894794.5A patent/CN115271096B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109063789A (en) * | 2018-09-07 | 2018-12-21 | 北京数字绿土科技有限公司 | The classification method of shaft tower and power line, device and system |
| WO2021008202A1 (en) * | 2019-07-16 | 2021-01-21 | 浙江大学 | Method for kernel support vector machine target classification based on millimeter-wave radar point cloud features |
| CN112513679A (en) * | 2020-04-15 | 2021-03-16 | 华为技术有限公司 | Target identification method and device |
| CN112837239A (en) * | 2021-02-01 | 2021-05-25 | 意诺科技有限公司 | Residual image eliminating method and device for millimeter wave radar and computer readable medium |
| CN113516052A (en) * | 2021-05-21 | 2021-10-19 | 同济大学 | Imaging millimeter wave radar point cloud target classification method based on machine learning |
| CN113763263A (en) * | 2021-07-27 | 2021-12-07 | 华能伊敏煤电有限责任公司 | Water mist tail gas noise treatment method based on point cloud tail gas filtering technology |
Non-Patent Citations (1)
| Title |
|---|
| 支持向量学习机在点云去噪中的应用;张琴等;《计算机技术与发展》;20110610(第06期);第85-88,94页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115271096A (en) | 2022-11-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113902897B (en) | Training of target detection model, target detection method, device, equipment and medium | |
| CN111444767B (en) | Pedestrian detection and tracking method based on laser radar | |
| JP2006350645A (en) | Object detection device and learning device for the same | |
| CN113378760A (en) | Training target detection model and method and device for detecting target | |
| CN113657274B (en) | Table generation method and device, electronic equipment and storage medium | |
| JPWO2018173846A1 (en) | Object detection apparatus, object detection method, and object detection program | |
| CN115271096B (en) | Point cloud processing and machine learning model training method and device and automatic driving vehicle | |
| CN111753623A (en) | Method, device and equipment for detecting moving object and storage medium | |
| CN113326786A (en) | Data processing method, device, equipment, vehicle and storage medium | |
| CN113947188A (en) | Training method of target detection network and vehicle detection method | |
| KR20230036960A (en) | Obstacle detection method and device, automatic driving vehicle, equipment and storage medium | |
| CN115471805A (en) | Point cloud processing and deep learning model training method and device and automatic driving vehicle | |
| CN114882198A (en) | Target determination method, device, equipment and medium | |
| CN114692778A (en) | Multi-modal sample set generation method, training method and device for intelligent inspection | |
| CN114581732A (en) | Image processing and model training method, device, equipment and storage medium | |
| CN113989721A (en) | Target detection method and training method and device of target detection model | |
| CN113657177A (en) | Data processing method and device, electronic equipment and storage medium | |
| JP6657934B2 (en) | Object detection device | |
| CN116246077A (en) | Security state detection method, device, equipment and medium | |
| CN113920273B (en) | Image processing method, device, electronic equipment and storage medium | |
| CN116152702A (en) | Point cloud label acquisition method and device, electronic equipment and automatic driving vehicle | |
| CN114708616A (en) | Obstacle avoidance method, device, equipment and storage medium | |
| CN113989300A (en) | Lane line segmentation method and device, electronic equipment and storage medium | |
| CN113887630A (en) | Image classification method and device, electronic equipment and storage medium | |
| CN113936158A (en) | Label matching method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |