CN109948684B - Quality inspection method, device and equipment for laser radar point cloud data labeling quality - Google Patents
Quality inspection method, device and equipment for laser radar point cloud data labeling quality Download PDFInfo
- Publication number
- CN109948684B CN109948684B CN201910184793.XA CN201910184793A CN109948684B CN 109948684 B CN109948684 B CN 109948684B CN 201910184793 A CN201910184793 A CN 201910184793A CN 109948684 B CN109948684 B CN 109948684B
- Authority
- CN
- China
- Prior art keywords
- point cloud
- cloud data
- classification
- under
- data
- 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
Images
Abstract
The invention discloses a quality inspection method and device for laser radar point cloud data marking quality and related equipment thereof. The quality inspection method comprises the following steps: acquiring point cloud data; respectively detecting the point cloud data according to N classification models obtained by pre-training to obtain detection classification results of the point cloud data under the N classification models; respectively comparing and evaluating detection classification results of the point cloud data under the N classification models with labeling data of the point cloud data to obtain N evaluation scores of the point cloud data; and performing quality inspection on the labeling quality of the point cloud data according to the N evaluation scores. According to the method, the quality inspection can be reasonably performed on the marking quality of the point cloud data through the evaluation scores of the point cloud data by calculating the N evaluation scores of the point cloud data and then performing the quality inspection on the marking quality of the point cloud data according to the evaluation scores, so that the range of data with suspected false marks and missed marks can be effectively narrowed, and the problem data can be automatically found out.
Description
Technical Field
The invention relates to the field of electronic data processing, in particular to a quality inspection method and device for marking quality of laser radar point cloud data, computer equipment and a computer readable storage medium.
Background
With the omnibearing development of social and economic and the continuous expansion of the demand for space information, in the current automatic driving perception technology, 3D (Three dimensional) laser radar point cloud data is generally used for perception, in order to enable the laser radar to acquire more structural information and to scan and process high-precision data, the laser radar point cloud data is used for calculation, and the point cloud data is marked as a key point in algorithm training. The quality of the data label directly affects the quality of the model.
In the related art, when quality inspection is performed on the labeling quality of the laser radar point cloud data, a manual quality inspector generally observes and judges whether a labeling error exists or not by naked eyes. However, such a manual quality inspection method may cause human errors, i.e., "false or missing marks" due to large data volume and human negligence, and is costly. Therefore, how to implement quality inspection on the labeling quality of point cloud data to effectively narrow the range of data with suspected false marks and missed marks and automatically find out problematic data becomes a problem to be solved urgently.
Disclosure of Invention
The object of the present invention is to solve at least to some extent one of the above mentioned technical problems.
Therefore, the first objective of the present invention is to provide a quality inspection method for laser radar point cloud data marking quality, which can reasonably perform quality inspection on the marking quality of the point cloud data through the evaluation score of the point cloud data, effectively reduce the range of "suspected false mark and missed mark" data, and realize automatic finding of problematic data, thereby reducing labor cost.
The second purpose of the invention is to provide a quality inspection device for laser radar point cloud data labeling quality.
A third object of the invention is to propose a computer device.
A fourth object of the invention is to propose a computer-readable storage medium.
In order to achieve the above object, a quality inspection method for laser radar point cloud data labeling quality provided by an embodiment of the first aspect of the present invention includes: acquiring point cloud data, wherein the point cloud data is marked point cloud data; respectively detecting the point cloud data according to N classification models obtained by pre-training to obtain detection classification results of the point cloud data under the N classification models, wherein N is a positive integer; comparing and evaluating detection classification results of the point cloud data under the N classification models with labeling data of the point cloud data respectively to obtain N evaluation scores of the point cloud data; and performing quality inspection on the labeling quality of the point cloud data according to the N evaluation scores.
According to the quality inspection method for the labeling quality of the laser radar point cloud data, the point cloud data can be obtained, then the point cloud data are respectively detected according to N classification models obtained through pre-training, detection classification results of the point cloud data under the N classification models are obtained, then the detection classification results of the point cloud data under the N classification models are respectively compared with the labeling data of the point cloud data to obtain N evaluation scores of the point cloud data, and finally the labeling quality of the point cloud data is subjected to quality inspection according to the N evaluation scores. The point cloud data are detected through the N classification models respectively, the obtained N detection results are compared and evaluated with the marking data of the point cloud data respectively to obtain N evaluation scores of the point cloud data, then the quality of the marking quality of the point cloud data can be reasonably checked according to the evaluation scores, the range of data which are suspected to be error marked and missed marked can be effectively narrowed, the problem data can be automatically found out, and the labor cost is reduced while the human errors can be reduced.
In order to achieve the above object, a quality inspection apparatus for laser radar point cloud data labeling quality according to an embodiment of a second aspect of the present invention includes: the data acquisition module is used for acquiring point cloud data, wherein the point cloud data is marked point cloud data; the data detection module is used for respectively detecting the point cloud data according to N classification models obtained by pre-training to obtain detection classification results of the point cloud data under the N classification models, wherein N is a positive integer; the data evaluation module is used for respectively comparing and evaluating the detection and classification results of the point cloud data under the N classification models with the labeling data of the point cloud data to obtain N evaluation scores of the point cloud data; and the quality inspection module is used for performing quality inspection on the labeling quality of the point cloud data according to the N evaluation scores.
According to the quality inspection device for the labeling quality of the laser radar point cloud data, the point cloud data can be obtained, then the point cloud data are respectively detected according to N classification models obtained through pre-training, detection classification results of the point cloud data under the N classification models are obtained, then the detection classification results of the point cloud data under the N classification models are respectively compared with the labeling data of the point cloud data to obtain N evaluation scores of the point cloud data, and finally the labeling quality of the point cloud data is subjected to quality inspection according to the N evaluation scores. The point cloud data are detected through the N classification models respectively, the obtained N detection results are compared with the marked data of the point cloud data respectively to obtain N evaluation scores of the point cloud data, quality inspection is carried out on the marked quality of the point cloud data according to the evaluation scores, and therefore quality inspection can be carried out on the marked quality of the point cloud data reasonably through the evaluation scores of the point cloud data, the range of data of suspected false marks and missed marks can be effectively reduced, automatic problem finding is achieved, and labor cost is reduced while human errors can be reduced.
To achieve the above object, a computer device according to a third embodiment of the present invention includes: the quality inspection method for the labeling quality of the laser radar point cloud data comprises a memory, a processor and a computer program which is stored on the memory and can run on the processor, wherein when the processor executes the computer program, the quality inspection method for the labeling quality of the laser radar point cloud data is realized.
In order to achieve the above object, a computer-readable storage medium is provided in an embodiment of a fourth aspect of the present invention, and the computer program is executed by a processor to implement the quality inspection method for labeling quality of lidar point cloud data according to the embodiment of the first aspect of the present invention.
Additional aspects and advantages of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.
Drawings
The foregoing and/or additional aspects and advantages of the present invention will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings of which:
fig. 1 is a flowchart of a quality inspection method for labeling quality of laser radar point cloud data according to an embodiment of the present invention.
Fig. 2 is a flowchart of a quality inspection method for laser radar point cloud data labeling quality according to an embodiment of the present invention.
Fig. 3 is a schematic structural diagram of a quality inspection apparatus for quality labeling of lidar point cloud data according to an embodiment of the present invention.
Fig. 4 is a schematic structural diagram of a quality inspection apparatus for quality labeling of lidar point cloud data according to another embodiment of the present invention.
FIG. 5 is a schematic diagram of a computer device according to one embodiment of the invention.
Detailed Description
Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the drawings are illustrative and intended to be illustrative of the invention and are not to be construed as limiting the invention.
The following explanation of terms in the embodiments of the present invention is presented:
the point cloud data refers to a set of vectors in a three-dimensional coordinate system, and may represent RGB (Red-Green-Blue, three primary colors) color, gray value, depth, segmentation result, and the like of a point in addition to representing geometric position information. The point cloud data is presented in a form equivalent to a 3D sample data, such as an image. The point cloud data in the invention is described by taking data collected by a 3D laser radar as an example. The examples are given solely for the purpose of illustration and are not to be construed as limitations of the present invention, as will be apparent to those of skill in the art.
In the related art, when quality inspection is performed on the labeling quality of point cloud data, a human quality inspector generally observes and judges whether a labeling error exists or not by naked eyes. However, such a manual quality inspection method may cause human errors, i.e., "false or missing marks" due to large data volume and human negligence, and is costly.
Therefore, the invention provides a quality inspection method and related equipment for the labeling quality of the point cloud data of the laser radar. Specifically, a quality inspection method, an apparatus, a computer device, and a computer-readable storage medium for laser radar point cloud data labeling quality according to embodiments of the present invention are described below with reference to the accompanying drawings.
Fig. 1 is a flowchart of a quality inspection method for labeling quality of laser radar point cloud data according to an embodiment of the present invention. It should be noted that the quality inspection method for laser radar point cloud data labeling quality in the embodiment of the present invention can be applied to the quality inspection device for laser radar point cloud data labeling quality in the embodiment of the present invention. The quality inspection device for the laser radar point cloud data marking quality can be configured on computer equipment. For example, the computer device may be a 3D scanning device, e.g., a hardware device such as a laser radar, stereo camera, etc.
As shown in fig. 1, the quality inspection method for labeling quality of the laser radar point cloud data may include:
and S110, acquiring point cloud data. It should be noted that, in the embodiment of the present invention, the point cloud data is labeled point cloud data. The point cloud data is marked manually or automatically by using an algorithm, which is not limited in the present invention.
It should be further noted that the point cloud data in the embodiment of the present invention may be point cloud data in a sample pool established in advance, for example, the point cloud data in the sample pool may be obtained by acquiring different images of various objects in advance through a scanning device, and then, the acquired point cloud data may be labeled, and the point cloud data and the labeled data thereof may be stored to establish the sample pool.
Optionally, the point cloud data of the embodiment of the present invention may be obtained by acquiring a field image of a certain target object by using a scanning device, and then labeling the point cloud data to obtain corresponding labeled data. Wherein the current scenario may include, but is not limited to: the method comprises the following steps of measuring a pile body, detecting a tunnel, surveying and mapping a highway, detecting a bridge, determining the right of land and the like, so that point cloud data under the scenes can be obtained.
It should be noted that the scanning device may be a laser radar, a Stereo Camera (Stereo Camera), a Time-of-flight Camera (Time-of-flight Camera) device, or the like.
And S120, respectively detecting the point cloud data according to N classification models obtained through pre-training to obtain detection classification results of the point cloud data under the N classification models. Wherein N is a positive integer.
Optionally, the point cloud data is detected according to the N classification models, and detection classification results of the point cloud data under the N classification models are obtained. That is, the point cloud data can be respectively input into the N classification models for detection, so as to obtain a detection classification result of the point cloud data under each classification model. For example, assuming that there are 10 classification models, the point cloud data may be respectively input into the 10 classification models, and each classification model performs classification detection on the point cloud data to obtain a corresponding detection classification result.
It should be noted that the N classification models are obtained by pre-training. In an embodiment of the present invention, the N classification models may be obtained by pre-training in the following manner:
s1) randomly extracting a specified number of samples from the marked point cloud data sample pool for multiple times;
s2) training a pre-created deep learning model according to the specified number of samples randomly extracted each time to obtain the N classification models; wherein the point cloud data does not include the samples randomly drawn each time for model training.
For example, based on the same labeled point cloud data sample pool, for example, 10 ten thousand frames of data, 1 ten thousand frames of data are randomly extracted each time. Training under the same model to obtain a plurality of models, for example, extracting 10 times in total, randomly extracting 1 ten thousand frames of data each time, and training the same model by using the randomly extracted data each time to finally obtain 10 models. It is understood that the difference of the N models includes randomness of data in addition to randomness in the model, for example, the model may have a function of identifying point cloud data and identifying whether there is an obstacle.
Thus, N classification models can be obtained through the pre-training of the above S1 and S2. It should be noted that, in order to ensure fairness and accuracy of quality inspection results of the labeling quality, in the embodiment of the present invention, reference is made to point cloud data subjected to model training without participating in quality inspection sequencing, that is, point cloud data subjected to quality inspection is data not subjected to model training.
S130, comparing and evaluating detection and classification results of the point cloud data under the N classification models with labeling data of the point cloud data respectively to obtain N evaluation scores of the point cloud data.
Optionally, a similarity measurement algorithm is adopted to calculate similarity coefficient values between detection classification results of the point cloud data under the N classification models and labeling data of the point cloud data, and then similarity coefficient values between detection classification results of the point cloud data under the N classification models and labeling data of the point cloud data are determined as N evaluation scores of the point cloud data, wherein the evaluation scores can be understood as similarity scores of the detection results and the labeling results. The similarity measurement algorithm may include, but is not limited to, a distance similarity calculation method (e.g., an euclidean distance algorithm), a cosine similarity calculation method, a jaccard distance algorithm, and the like, and is not limited specifically.
And S140, calculating the labeling quality of the point cloud data according to the N evaluation scores for quality inspection.
Optionally, according to the N evaluation scores, calculating a mean value and a variance of the point cloud data scores, then determining whether the mean value of the point cloud data scores is smaller than a first threshold, and determining whether the variance of the point cloud data scores is smaller than a second threshold, if the mean value is smaller than the first threshold and the variance is smaller than the second threshold, determining that the point cloud data is labeled as suspected false label and missed label. It should be noted that, in the embodiment of the present invention, the first threshold and the second threshold may be preset and need to be determined according to previous visualization.
According to the quality inspection method for the labeling quality of the laser radar point cloud data, the point cloud data can be obtained, then the point cloud data are respectively detected according to N classification models obtained through pre-training, detection classification results of the point cloud data under the N classification models are obtained, then the detection classification results of the point cloud data under the N classification models are respectively compared with the labeling data of the point cloud data to obtain N evaluation scores of the point cloud data, and finally the labeling quality of the point cloud data is subjected to quality inspection according to the N evaluation scores. The point cloud data are detected through the N classification models respectively, the obtained N detection results are compared with the marked data of the point cloud data respectively to obtain N evaluation scores of the point cloud data, quality inspection is carried out on the marked quality of the point cloud data according to the evaluation scores, and therefore quality inspection can be carried out on the marked quality of the point cloud data reasonably through the evaluation scores of the point cloud data, the range of data of suspected false marks and missed marks can be effectively reduced, automatic problem finding is achieved, and labor cost is reduced while human errors can be reduced.
Fig. 2 is a flowchart of a quality inspection method for laser radar point cloud data labeling quality according to an embodiment of the present invention. It should be noted that each frame of point cloud data may have a plurality of obstacles, and in order to examine the difficulty of the obstacles in one frame of point cloud data, the difficulty of each obstacle in each frame needs to be considered at the same time. Specifically, as shown in fig. 2, the quality inspection method for labeling quality of the laser radar point cloud data may include:
s210, point cloud data is obtained, wherein the point cloud data is marked point cloud data.
And S220, respectively detecting the point cloud data according to the N classification models obtained through pre-training to obtain detection classification results of each obstacle in the point cloud data under the N classification models.
Optionally, the point cloud data is detected according to the N classification models, and a detection classification result of each obstacle in the point cloud data under the N classification models is obtained. Wherein N is a positive integer. It should be noted that a frame of point cloud data may be composed of 2 parts, one part is a background of no interest, such as the ground, lawn, vegetation, etc., and the other part is a foreground of interest, such as pedestrians, vehicles, etc. These foreground objects may be called obstacles and are the targets to be detected.
That is to say, each classification model is used for carrying out classification detection on the point cloud data to obtain a detection classification result of each obstacle in the point cloud data, and each obstacle can correspondingly obtain N detection classification results due to N classification models.
And S230, calculating a detection classification result of each obstacle in the point cloud data under the ith classification model, and respectively calculating a similarity coefficient value between the detection classification result and the labeling data of each obstacle.
In the embodiment of the invention, a similarity measurement algorithm can be adopted to calculate the similarity coefficient value between the detection classification result of each obstacle in the point cloud data under the ith classification model and the labeling data of each obstacle, wherein i is more than or equal to 1 and less than or equal to N.
For example, taking a similarity metric algorithm as a jaccard distance algorithm as an example, in order to examine the difficulty of the obstacle in each frame of data, the difficulty of each obstacle in each frame needs to be considered at the same time, a detection value (i.e., a detection classification result) and a labeling value of each obstacle in each frame can be compared by using the similarity metric algorithm, so that a jaccard index (i.e., a similarity coefficient value) can be obtained, the jaccard index value of each obstacle represents the similarity degree between the detection result and the labeling result of the obstacle, and the larger the jaccard index value, the larger the coincidence degree between the detection result and the labeling value of the point cloud data is.
As an embodiment of a possible implementation manner, taking a similarity measurement algorithm as an euclidean distance algorithm as an example, in order to examine the difficulty of an obstacle in each frame of data, the difficulty of each obstacle in each frame needs to be considered at the same time, and a detection value (i.e., a detection classification result) and a label value of each obstacle in each frame can be compared by using the similarity measurement algorithm, so as to obtain a euclidean distance value, where the smaller the obtained euclidean distance value is, the greater the similarity between the detection value of the detection classification result and the label value of the point cloud data is, the greater the obtained euclidean distance value is, and the smaller the similarity between the detection value of the detection classification result and the label value of the point cloud data is.
S240, determining the detection classification result of each obstacle in the point cloud data under the ith classification model and the similarity coefficient value between the detection classification result and the labeling data of each obstacle respectively as the evaluation score of each obstacle under the ith classification model.
That is to say, the detection classification result of each obstacle in the point cloud data under the ith classification model is compared with the labeling data of each obstacle to obtain a similarity coefficient value, wherein the similarity coefficient value is the evaluation score of each obstacle under the ith classification model.
And S250, acquiring the evaluation score of the point cloud data under the ith classification model according to the evaluation score of each obstacle under the ith classification model.
Optionally, the evaluation scores of all the obstacles in the point cloud data under the ith classification model are ranked, then a quantile is selected from the ranked evaluation scores of all the obstacles under the ith classification model, and the selected quantile is used for determining the evaluation score of the point cloud data under the ith classification model. In the embodiment of the present invention, the quantile may be a median, and may also be other different quantiles, for example, a quartile, a decile, and the like.
That is, after the similarity coefficient values of all the obstacles in a frame of labeled data are obtained, the similarity coefficient values are sorted, and then the quantile of the sorting result is taken as the evaluation score of the similarity degree between the detection result and the labeled result of the frame under the current model. As a possible implementation example, after the similarity coefficient values of all obstacles in a frame of labeled data are obtained, the similarity coefficient values are sorted, and an average value of the sorting results can also be taken as an evaluation score of the frame under the current i-th classification model.
In the embodiment of the invention, the N evaluation scores may be sorted from large to small, and the sorted median is taken as the N evaluation scores of the point cloud data. Optionally, the N evaluation scores may be averaged or weighted averaged, and the obtained value is used as the N evaluation scores of the point cloud data.
That is, the evaluation score of each obstacle in 1 frame of point cloud data is determined (in a model), and thus N evaluation scores of the point cloud data can be obtained by repeating the above steps N times.
And S260, performing quality inspection on the labeling quality of the point cloud data according to the N evaluation scores.
Optionally, in an embodiment of the present invention, according to the N evaluation scores, then a mean and a variance of the point cloud data scores are calculated, then it is determined whether the mean of the point cloud data scores is smaller than a first threshold and whether the variance of the point cloud data scores is smaller than a second threshold, and if the mean is smaller than the first threshold and the variance is smaller than the second threshold, it is determined that the point cloud data is labeled as a suspected false label or a missed label.
For example, taking 10 evaluation scores as an example, after the 10 evaluation scores are obtained, the mean and variance of the point cloud data scores are calculated by using a formula, and then whether the mean of the point cloud data scores is smaller than a first threshold value and whether the variance of the point cloud data scores is smaller than a second threshold value are judged. And if the mean value is smaller than a first threshold value and the variance is smaller than a second threshold value, judging that the point cloud data is marked as suspected false marks and missed marks, and performing automatic quality inspection after determining the suspected false marks and the missed marks.
Specifically, all the score means are sorted, wherein the data with lower score mean and lower variance (i.e. all models cannot detect) is "suspected false bid, missed bid".
According to the quality inspection method for the labeling quality of the laser radar point cloud data, point cloud data are obtained, the point cloud data are detected according to N classification models respectively, the detection classification results of each obstacle in the point cloud data under the N classification models are obtained, the similarity coefficient value between the detection classification result of each obstacle in the point cloud data under the ith classification model and the labeling data of each obstacle is calculated, the detection classification result of each obstacle in the point cloud data under the ith classification model and the similarity coefficient value between each detection classification result of each obstacle in the point cloud data and the labeling data of each obstacle are determined, the evaluation score of each obstacle in the ith classification model is determined, and the evaluation score of the point cloud data under the ith classification model is obtained according to the evaluation score of each obstacle in the ith classification model, and finally, performing quality inspection on the labeling quality of the point cloud data according to the N evaluation scores. The quality of the point cloud data can be reasonably checked through the evaluation score of the point cloud data, the range of data which are suspected of error labeling and label missing can be effectively narrowed, and problematic data can be automatically found out.
The quality inspection method for the laser radar point cloud data marking quality provided by the embodiment of the invention is also suitable for the quality inspection device for the laser radar point cloud data marking quality provided by the embodiment of the invention, and detailed description is not provided in the embodiment of the invention. Fig. 3 is a schematic structural diagram of a quality inspection apparatus for quality labeling of lidar point cloud data according to an embodiment of the present invention.
As shown in fig. 3, the quality inspection device 300 for labeling quality of the point lidar cloud data includes: a data acquisition module 310, a data detection module 320, a data evaluation module 330, and a quality inspection module 340, wherein:
the data obtaining module 310 is configured to obtain point cloud data, where the point cloud data is labeled point cloud data.
The data detection module 320 is configured to detect the point cloud data according to N classification models obtained through pre-training, respectively, to obtain detection classification results of the point cloud data under the N classification models, where N is a positive integer.
The data evaluation module 330 is configured to compare and evaluate the detection classification results of the point cloud data under the N classification models with the labeling data of the point cloud data, respectively, to obtain N evaluation scores of the point cloud data.
The quality inspection module 340 is configured to perform quality inspection on the labeling quality of the point cloud data according to the N evaluation scores.
According to the quality inspection device for the labeling quality of the laser radar point cloud data, the point cloud data can be obtained, then the point cloud data are respectively detected according to N classification models obtained through pre-training, detection classification results of the point cloud data under the N classification models are obtained, then the detection classification results of the point cloud data under the N classification models are respectively compared with the labeling data of the point cloud data to obtain N evaluation scores of the point cloud data, and finally the labeling quality of the point cloud data is subjected to quality inspection according to the N evaluation scores. The point cloud data are detected through the N classification models respectively, the obtained N detection results are compared with the marked data of the point cloud data respectively to obtain N evaluation scores of the point cloud data, quality inspection is carried out on the marked quality of the point cloud data according to the evaluation scores, and therefore quality inspection can be carried out on the marked quality of the point cloud data reasonably through the evaluation scores of the point cloud data, the range of data of suspected false marks and missed marks can be effectively reduced, automatic problem finding is achieved, and labor cost is reduced while human errors can be reduced.
It should be noted that the N classification models are obtained by pre-training. Optionally, in an embodiment of the present invention, as shown in fig. 4, the apparatus for quality inspection of laser radar point cloud data annotation further includes: the model training module 350 is used for pre-training to obtain N classification models; wherein, the model training module 350 is specifically configured to: randomly extracting a specified number of samples from a marked point cloud data sample pool for multiple times; training a pre-created deep learning model according to the specified number of samples randomly extracted each time to obtain N classification models; wherein the point cloud data does not contain the samples randomly drawn each time for model training.
It should be noted that each frame of point cloud data may have a plurality of obstacles, and in order to examine the difficulty of the obstacles in one frame of point cloud data, the difficulty of each obstacle in each frame needs to be considered at the same time. Specifically, in an embodiment of the present invention, the data detecting module 320 is specifically configured to: and respectively detecting the point cloud data according to the N classification models to obtain the detection classification result of each obstacle in the point cloud data under the N classification models.
In one embodiment of the invention, the data evaluation module 330 includes: the calculation unit is used for calculating a detection classification result of each obstacle in the point cloud data under the ith classification model and a similarity coefficient value between each detection classification result and the labeled data of each obstacle, wherein i is more than or equal to 1 and less than or equal to N; the determining unit is used for determining a detection classification result of each obstacle in the point cloud data under the ith classification model and a similarity coefficient value between the detection classification result and the labeling data of each obstacle respectively as an evaluation score of each obstacle under the ith classification model; and the score acquisition unit is used for acquiring the evaluation score of the point cloud data under the ith classification model according to the evaluation score of each obstacle under the ith classification model.
In an embodiment of the present invention, the score obtaining unit is specifically configured to: ranking the evaluation scores of all obstacles in the point cloud data under the ith classification model; and selecting quantiles from the evaluation scores of all the sorted obstacles under the ith classification model, and determining the evaluation score of the point cloud data under the ith classification model according to the selected quantiles.
In an embodiment of the present invention, the quality inspection module 340 is specifically configured to calculate a mean and a variance of the point cloud data scores according to the N evaluation scores; judging whether the mean value of the point cloud data scores is smaller than a first threshold value or not; judging whether the variance of the point cloud data scores is smaller than a second threshold value; and when the mean value is smaller than a first threshold value and the variance is smaller than a second threshold value, judging that the point cloud data is marked as suspected false mark and missed mark.
In order to implement the above embodiments, the present invention further provides a computer device.
FIG. 5 is a schematic diagram of a computer device according to one embodiment of the invention. As shown in fig. 5, the server 500 may include: the storage 510, the processor 520, and the computer program 530 stored in the storage 510 and operable on the processor 520, when the processor 520 executes the program, the method for quality inspection of the labeling quality of the lidar point cloud data according to any of the above embodiments of the present invention is implemented.
In order to implement the above embodiments, the present invention further provides a computer program, and when the computer program is executed by a processor, the quality inspection method for the labeling quality of the laser radar point cloud data is implemented.
In the description herein, references to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, various embodiments or examples and features of different embodiments or examples described in this specification can be combined and combined by one skilled in the art without contradiction.
Any process or method descriptions in flow charts or otherwise described herein may be understood as representing modules, segments, or portions of code which include one or more executable instructions for implementing specific logical functions or steps of the process, and alternate implementations are included within the scope of the preferred embodiment of the present invention in which functions may be executed out of order from that shown or discussed, including substantially concurrently or in reverse order, depending on the functionality involved, as would be understood by those reasonably skilled in the art of the present invention.
The logic and/or steps represented in the flowcharts or otherwise described herein, e.g., an ordered listing of executable instructions that can be considered to implement logical functions, can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For the purposes of this description, a "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of the computer-readable medium would include the following: an electrical connection (electronic device) having one or more wires, a portable computer diskette (magnetic device), a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium could even be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, via for instance optical scanning of the paper or other medium, then compiled, interpreted or otherwise processed in a suitable manner if necessary, and then stored in a computer memory.
It should be understood that portions of the present invention may be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, the various steps or methods may be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one or combination of the following techniques, which are known in the art, may be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application specific integrated circuit having an appropriate combinational logic gate circuit, a Programmable Gate Array (PGA), a Field Programmable Gate Array (FPGA), or the like.
It will be understood by those skilled in the art that all or part of the steps carried by the method for implementing the above embodiments may be implemented by hardware related to instructions of a program, which may be stored in a computer readable storage medium, and when the program is executed, the program includes one or a combination of the steps of the method embodiments.
In addition, functional units in the embodiments of the present invention may be integrated into one processing module, or each unit may exist alone physically, or two or more units are integrated into one module. The integrated module can be realized in a hardware mode, and can also be realized in a software functional module mode. The integrated module, if implemented in the form of a software functional module and sold or used as a stand-alone product, may also be stored in a computer readable storage medium.
The storage medium mentioned above may be a read-only memory, a magnetic or optical disk, etc. Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention, and that variations, modifications, substitutions and alterations can be made to the above embodiments by those of ordinary skill in the art within the scope of the present invention.
Claims (10)
1. A quality inspection method for laser radar point cloud data labeling quality is characterized by comprising the following steps:
acquiring point cloud data, wherein the point cloud data is marked point cloud data;
respectively detecting the point cloud data according to N classification models obtained by pre-training to obtain detection classification results of the point cloud data under the N classification models, wherein N is a positive integer;
comparing and evaluating detection classification results of the point cloud data under the N classification models with labeling data of the point cloud data respectively to obtain N evaluation scores of the point cloud data, calculating similarity coefficient values between the detection classification results of the point cloud data under the N classification models and the labeling data of the point cloud data respectively, and determining the similarity coefficient values between the detection classification results of the point cloud data under the N classification models and the labeling data of the point cloud data respectively as the N evaluation scores of the point cloud data;
performing quality inspection on the labeling quality of the point cloud data according to the N evaluation scores;
the method for detecting the point cloud data according to N classification models obtained through pre-training to obtain detection classification results of the point cloud data under the N classification models comprises the following steps:
respectively detecting the point cloud data according to the N classification models to obtain detection classification results of each obstacle in the point cloud data under the N classification models;
the step of comparing and evaluating the detection classification results of the point cloud data under the N classification models with the labeling data of the point cloud data to obtain N evaluation scores of the point cloud data comprises the following steps:
calculating a detection classification result of each obstacle in the point cloud data under an ith classification model, and respectively calculating a similarity coefficient value between the detection classification result and the labeled data of each obstacle, wherein i is more than or equal to 1 and less than or equal to N;
determining a detection classification result of each obstacle in the point cloud data under the ith classification model and a similarity coefficient value between the detection classification result and the labeled data of each obstacle respectively as an evaluation score of each obstacle under the ith classification model;
and obtaining the evaluation score of the point cloud data under the ith classification model according to the evaluation score of each obstacle under the ith classification model.
2. The method of claim 1, wherein the N classification models are pre-trained by:
randomly extracting a specified number of samples from a marked point cloud data sample pool for multiple times;
training a pre-created deep learning model according to the specified number of samples randomly extracted each time to obtain the N classification models; wherein the point cloud data does not include the samples randomly drawn each time for model training.
3. The method of claim 1, wherein obtaining the evaluation score of the point cloud data under the ith classification model according to the evaluation score of each obstacle under the ith classification model comprises:
ranking the evaluation scores of all obstacles in the point cloud data under the ith classification model;
selecting quantiles from the evaluation scores of all the sorted obstacles under the ith classification model, and determining the evaluation score of the point cloud data under the ith classification model according to the selected quantiles.
4. The method of claim 1, wherein quality testing the labeling quality of the point cloud data according to the N evaluation scores comprises:
calculating the mean and variance of the point cloud data scores according to the N evaluation scores;
judging whether the mean value of the point cloud data scores is smaller than a first threshold value or not;
judging whether the variance of the point cloud data scores is smaller than a second threshold value;
and if the mean value is smaller than the first threshold value and the variance is smaller than a second threshold value, judging that the point cloud data is marked as suspected false mark and missed mark.
5. A quality inspection device for laser radar point cloud data labeling quality is characterized by comprising:
the data acquisition module is used for acquiring point cloud data, wherein the point cloud data is marked point cloud data;
the data detection module is used for respectively detecting the point cloud data according to N classification models obtained by pre-training to obtain detection classification results of the point cloud data under the N classification models, wherein N is a positive integer;
the data evaluation module is used for respectively comparing and evaluating the detection classification results of the point cloud data under the N classification models with the labeled data of the point cloud data to obtain N evaluation scores of the point cloud data, calculating the similarity coefficient values between the detection classification results of the point cloud data under the N classification models and the labeled data of the point cloud data, and determining the similarity coefficient values between the detection classification results of the point cloud data under the N classification models and the labeled data of the point cloud data as the N evaluation scores of the point cloud data;
the quality inspection module is used for performing quality inspection on the labeling quality of the point cloud data according to the N evaluation scores;
the data detection module is specifically configured to:
respectively detecting the point cloud data according to the N classification models to obtain detection classification results of each obstacle in the point cloud data under the N classification models;
the calculation unit is used for calculating a detection classification result of each obstacle in the point cloud data under the ith classification model and a similarity coefficient value between the detection classification result and the labeled data of each obstacle, wherein i is more than or equal to 1 and less than or equal to N;
a determining unit, configured to determine, as an evaluation score of each obstacle under the ith classification model, a detection classification result of each obstacle under the ith classification model in the point cloud data and a similarity coefficient value between the detection classification result and the labeling data of each obstacle respectively;
and the score acquisition unit is used for acquiring the evaluation score of the point cloud data under the ith classification model according to the evaluation score of each obstacle under the ith classification model.
6. The apparatus of claim 5, further comprising:
the model training module is used for obtaining the N classification models through pre-training;
wherein the model training module is specifically configured to:
randomly extracting a specified number of samples from a marked point cloud data sample pool for multiple times;
training a pre-created deep learning model according to the specified number of samples randomly extracted each time to obtain the N classification models; wherein the point cloud data does not include the samples randomly drawn each time for model training.
7. The apparatus according to claim 5, wherein the score obtaining unit is specifically configured to:
ranking the evaluation scores of all obstacles in the point cloud data under the ith classification model;
selecting quantiles from the evaluation scores of all the sorted obstacles under the ith classification model, and determining the evaluation score of the point cloud data under the ith classification model according to the selected quantiles.
8. The apparatus of claim 5, wherein the quality inspection module is specifically configured to:
calculating the mean and variance of the point cloud data scores according to the N evaluation scores;
judging whether the mean value of the point cloud data scores is smaller than a first threshold value or not;
judging whether the variance of the point cloud data scores is smaller than a second threshold value;
and when the mean value is smaller than the first threshold value and the variance is smaller than a second threshold value, judging that the point cloud data is marked as suspected false mark and missed mark.
9. A computer device, comprising: the device comprises a memory, a processor and a computer program which is stored on the memory and can run on the processor, wherein when the processor executes the computer program, the quality inspection method for the labeling quality of the laser radar point cloud data is realized according to any one of claims 1 to 4.
10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, implements a quality inspection method for quality labeling of lidar point cloud data according to any of claims 1 to 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910184793.XA CN109948684B (en) | 2019-03-12 | 2019-03-12 | Quality inspection method, device and equipment for laser radar point cloud data labeling quality |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910184793.XA CN109948684B (en) | 2019-03-12 | 2019-03-12 | Quality inspection method, device and equipment for laser radar point cloud data labeling quality |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109948684A CN109948684A (en) | 2019-06-28 |
| CN109948684B true CN109948684B (en) | 2022-01-18 |
Family
ID=67009742
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910184793.XA Active CN109948684B (en) | 2019-03-12 | 2019-03-12 | Quality inspection method, device and equipment for laser radar point cloud data labeling quality |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109948684B (en) |
Families Citing this family (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110610004A (en) * | 2019-09-03 | 2019-12-24 | 深圳追一科技有限公司 | Method and device for detecting labeling quality, computer equipment and storage medium |
| CN111259980B (en) * | 2020-02-10 | 2023-10-03 | 北京小马慧行科技有限公司 | Method and device for processing annotation data |
| CN112685765A (en) * | 2020-03-25 | 2021-04-20 | 华控清交信息科技(北京)有限公司 | Data quality evaluation method and device for data quality evaluation |
| CN113591888A (en) * | 2020-04-30 | 2021-11-02 | 上海禾赛科技有限公司 | Point cloud data labeling network system and method for laser radar |
| CN112100497B (en) * | 2020-09-14 | 2021-10-19 | 北京嘀嘀无限科技发展有限公司 | Data processing method and device, electronic equipment and readable storage medium |
| CN112200273B (en) * | 2020-12-07 | 2021-05-07 | 长沙海信智能系统研究院有限公司 | Data annotation method, device, equipment and computer storage medium |
| CN112698421A (en) * | 2020-12-11 | 2021-04-23 | 北京百度网讯科技有限公司 | Evaluation method, device, equipment and storage medium for obstacle detection |
| CN112801200B (en) * | 2021-02-07 | 2024-02-20 | 文远鄂行(湖北)出行科技有限公司 | Data packet screening method, device, equipment and storage medium |
| CN113469020A (en) * | 2021-06-29 | 2021-10-01 | 苏州一径科技有限公司 | Target detection model evaluation method based on clustering |
| CN113642416A (en) * | 2021-07-20 | 2021-11-12 | 武汉光庭信息技术股份有限公司 | Test cloud platform for AI (Artificial intelligence) annotation and AI annotation test method |
| CN116246273B (en) * | 2023-03-07 | 2024-03-22 | 广州市易鸿智能装备有限公司 | Image annotation consistency evaluation method and device, electronic equipment and storage medium |
| CN116413740B (en) * | 2023-06-09 | 2023-09-05 | 广汽埃安新能源汽车股份有限公司 | Laser radar point cloud ground detection method and device |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109298786A (en) * | 2018-09-13 | 2019-02-01 | 北京旷视科技有限公司 | Mark accuracy rate appraisal procedure and device |
| CN109344804A (en) * | 2018-10-30 | 2019-02-15 | 百度在线网络技术（北京）有限公司 | A kind of recognition methods of laser point cloud data, device, equipment and medium |
Family Cites Families (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| GB2543749A (en) * | 2015-10-21 | 2017-05-03 | Nokia Technologies Oy | 3D scene rendering |
| CN108509969B (en) * | 2017-09-06 | 2021-11-09 | 腾讯科技（深圳）有限公司 | Data labeling method and terminal |
| CN107730585A (en) * | 2017-11-06 | 2018-02-23 | 济南市市政工程设计研究院（集团）有限责任公司 | A kind of landform threedimensional model generation method and system |
| CN108647264B (en) * | 2018-04-28 | 2020-10-13 | 北京邮电大学 | Automatic image annotation method and device based on support vector machine |
| CN108898162B (en) * | 2018-06-08 | 2021-03-30 | 东软集团股份有限公司 | Data annotation method, device and equipment and computer readable storage medium |
| CN108985171B (en) * | 2018-06-15 | 2023-04-07 | 上海仙途智能科技有限公司 | Motion state estimation method and motion state estimation device |
| CN108830329B (en) * | 2018-06-22 | 2020-07-10 | 北京字节跳动网络技术有限公司 | Picture processing method and device |
| CN109144988B (en) * | 2018-08-07 | 2021-02-23 | 东软集团股份有限公司 | Abnormal data detection method and device |
-
2019
- 2019-03-12 CN CN201910184793.XA patent/CN109948684B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109298786A (en) * | 2018-09-13 | 2019-02-01 | 北京旷视科技有限公司 | Mark accuracy rate appraisal procedure and device |
| CN109344804A (en) * | 2018-10-30 | 2019-02-15 | 百度在线网络技术（北京）有限公司 | A kind of recognition methods of laser point cloud data, device, equipment and medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109948684A (en) | 2019-06-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109948684B (en) | Quality inspection method, device and equipment for laser radar point cloud data labeling quality | |
| CN109948683B (en) | Method and device for dividing difficulty degree of point cloud data and related equipment thereof | |
| CN110135307B (en) | Traffic sign detection method and device based on attention mechanism | |
| EP3220353A1 (en) | Image processing apparatus, image processing method, and recording medium | |
| CN111126393A (en) | Vehicle appearance refitting judgment method and device, computer equipment and storage medium | |
| CN112669290A (en) | Image comparison method and device | |
| Somawirata et al. | Road detection based on the color space and cluster connecting | |
| JP6698191B1 (en) | Road marking failure detection device, road marking failure detection method, and road marking failure detection program | |
| CN113554645B (en) | Industrial anomaly detection method and device based on WGAN | |
| CN116824516B (en) | Road construction safety monitoring and management system | |
| CN112183382A (en) | Unmanned traffic light detection and classification method and device | |
| CN115713750B (en) | Lane line detection method and device, electronic equipment and storage medium | |
| CN117132768A (en) | License plate and face detection and desensitization method and device, electronic equipment and storage medium | |
| Li et al. | Road Damage Evaluation via Stereo Camera and Deep Learning Neural Network | |
| CN113674361B (en) | Vehicle-mounted all-round-looking calibration implementation method and system | |
| CN114140671A (en) | High-resolution chip image classification method and device based on multi-scale fusion | |
| CN115018787A (en) | Anomaly detection method and system based on gradient enhancement | |
| CN112700446A (en) | Algorithm model training method and device for industrial quality inspection | |
| CN113468931A (en) | Data processing method and device, electronic equipment and storage medium | |
| CN115063594B (en) | Feature extraction method and device based on automatic driving | |
| US20220402504A1 (en) | Methods and Systems for Generating Ground Truth Data | |
| CN113419075B (en) | Ship speed measuring method, system, device and medium based on binocular vision | |
| CN117474915B (en) | Abnormality detection method, electronic equipment and storage medium | |
| CN115641567B (en) | Target object detection method and device for vehicle, vehicle and medium | |
| Zuckerman et al. | Distance Estimation to Image Objects Using Adapted Scale |
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 | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20211012 Address after: 105 / F, building 1, No. 10, Shangdi 10th Street, Haidian District, Beijing 100085 Applicant after: Apollo Intelligent Technology (Beijing) Co.,Ltd. Address before: 100085 Baidu Building, 10 Shangdi Tenth Street, Haidian District, Beijing Applicant before: BAIDU ONLINE NETWORK TECHNOLOGY (BEIJING) Co.,Ltd. |
|
| TA01 | Transfer of patent application right | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |