CN111598885B

CN111598885B - Automatic visibility grade marking method for highway foggy pictures

Info

Publication number: CN111598885B
Application number: CN202010437766.1A
Authority: CN
Inventors: 杨卓敏; 张森; 李�杰; 尤冬海; 张慧辰
Original assignee: Traffic Management Research Institute of Ministry of Public Security
Current assignee: Traffic Management Research Institute of Ministry of Public Security
Priority date: 2020-05-21
Filing date: 2020-05-21
Publication date: 2022-02-11
Anticipated expiration: 2040-05-21
Also published as: CN111598885A

Abstract

The invention provides an automatic marking method for the visibility grade of a highway foggy picture, which comprises the following steps: step S1, testing a plurality of known foggy detection algorithms by using the labeled foggy picture test sample set to obtain an identification accuracy matrix A of the plurality of known foggy detection algorithms; step S2, identifying the to-be-labeled group fog pictures by utilizing the plurality of known group fog detection algorithms to obtain a corresponding identification result matrix P; step S3, obtaining a voting result matrix F by using the formula F = a × P; step S4, calculating a maximum vote value in the voting result matrix; step S5, automatically labeling the visibility grade corresponding to the maximum voting value with a group fog picture to be labeled; and S6, repeating the steps S2-S5, and automatically marking the visibility levels of all the group fog pictures to be marked in the group fog pictures to be marked. The invention avoids the problem of inconsistent labeling results caused by manual labeling, and has higher labeling efficiency.

Description

Automatic visibility grade marking method for highway foggy pictures

Technical Field

The invention belongs to the technical field of image recognition, and particularly relates to an automatic highway foggy picture visibility grade marking method.

Background

The field of deep learning is rapidly developed recently, and particularly, the application of the deep learning technology to image recognition is mature, and the group fog recognition based on the monitoring image of the expressway also starts to apply the deep learning technology. The key point of the deep learning technology lies in the collection and labeling of a large number of training sample pictures, the difficulty of highway group fog identification lies in the labeling of the visibility grade of the sample pictures, and the visibility value of the visibility detector is difficult to match with the monitoring image due to the following reasons:

1. visibility detector deployment density is lower

On average, the visibility detectors on the highway are about 10 kilometers 1 set, the number of intact equipment is less, and the monitoring cameras on the highway are 1 kilometer 1 set on average.

2. The visibility detector and the monitoring camera are not deployed at the same point and have a considerable distance

3. Problem of roadside occlusion

Visibility detector generally installs the highway roadside, and the mounting height is only about 1.5 meters, is difficult to avoid being sheltered from by trees, sign sighting rod etc. of roadside, influences the accuracy of visibility.

4. Distinction between point monitoring and area monitoring

The visibility detector is used for point monitoring, the obtained visibility value only represents one point, and the monitoring image is used for surface monitoring and represents the visibility condition of the whole road section within an effective monitoring distance (such as 100 meters). The cluster fog is affected by the microclimate environment of local areas, and in the local range of tens to hundreds of meters in the big fog, the fog with higher concentration and lower visibility appears. Due to the regional nature of the cloud, it is likely that the visibility detector area is not covered.

Therefore, the labeling of the visibility grade of the foggy sample picture cannot only depend on the visibility value obtained by the visibility detector, and other labeling methods are necessary.

In the aspect of visual application, deep learning mainly comprises three types of target classification, target detection and semantic segmentation, and in the aspect of group fog recognition, only the target classification is applicable, and continuous visibility values cannot be obtained.

Referring to the standards of the police and meteorological departments, the images of the cloud can be classified into six levels:

(1) visibility level one

Representing visibility below 50 meters.

(2) Visibility second level

Representing a visibility of 50 to 100 meters.

(3) Visibility is tertiary

Representing a visibility of 100 to 200 meters.

(4) Visibility four-level

Representing a visibility of 200 to 500 meters.

(5) Visibility five-level

Representing a visibility of 500 to 1000 meters.

(6) Fog-free

Visibility above 1000 meters is considered fog-free.

The police and the traffic department mainly perform corresponding road management and control according to the risk level of the group fog, and the group fog visibility level is distinguished by adopting a deep learning algorithm, so that the service requirement can be met.

However, the visibility level detection algorithm based on deep learning needs a large number of labeled sample pictures, the manual labeling is high in labor cost, the judgment of the labeling personnel on the visibility is very subjective, and the condition that the labeling results are inconsistent can occur even if the same labeling personnel performs labeling twice. There is a need for an automated labeling method with high objectivity.

Disclosure of Invention

The invention aims to overcome the defects in the prior art, provides an automatic labeling method for the visibility grade of a highway group fog picture, avoids the problem of inconsistent labeling results caused by manual labeling, adopts a plurality of group fog detection algorithms to vote, selects the visibility grade with the most voted tickets as a labeling result, and realizes scientific automatic labeling. The technical scheme adopted by the invention is as follows:

a highway foggy image visibility grade automatic labeling method comprises the following steps:

step S1, testing a plurality of known foggy detection algorithms by using the labeled foggy picture test sample set to obtain an identification accuracy matrix A of the plurality of known foggy detection algorithms;

step S2, identifying the to-be-labeled group fog pictures by utilizing the plurality of known group fog detection algorithms to obtain a corresponding identification result matrix P;

step S3, obtaining a voting result matrix F according to the formula F ═ a × P;

step S4, calculating a maximum vote value in the voting result matrix;

step S5, automatically labeling the visibility grade corresponding to the maximum voting value with a group fog picture to be labeled;

and S6, repeating the steps S2-S5, and automatically marking the visibility levels of all the group fog pictures to be marked in the group fog pictures to be marked.

Further, in step S5, if there are a plurality of equal maximum vote values or the maximum vote values are smaller than the set threshold value, the labeling is discarded.

Further, step S1 specifically includes:

step S101, selecting a plurality of known cloud detection algorithms; selecting algorithms with different principles during selection;

step S102, preparing a marked group fog picture test sample set;

the cluster fog picture test samples in the cluster fog picture test sample set adopt cluster fog pictures which are verified with the visibility detector data;

and step S103, testing a plurality of known cluster fog detection algorithms by using the labeled cluster fog picture test sample set to obtain a corresponding accuracy matrix A.

Further, the visibility grade is divided into five grades;

(1) visibility first, representing visibility below 50 meters;

(2) visibility level two, representing visibility 50 to 100 meters;

(3) visibility three-level, representing visibility 100 to 200 meters;

(4) visibility four, representing visibility 200 to 500 meters;

(5) visibility grade five, representing visibility 500 to 1000 meters.

The invention has the advantages that:

1) compared with manual labeling, the labeling efficiency is greatly improved.

2) Compared with the uncertainty of manual labeling, the method provided by the invention identifies through each group fog detection algorithm, and can eliminate the uncertainty of labeling when the results tend to be consistent for many times.

3) The algorithm recognition accuracy is used as the voting weight for counting the votes, and both the objectivity and the accuracy are improved.

Drawings

FIG. 1 is a flow chart of the method of the present invention.

Detailed Description

The invention is further illustrated by the following specific figures and examples.

The embodiment of the invention provides an automatic marking method for the visibility grade of a highway foggy picture, which comprises the following steps:

step S1 specifically includes:

step S101, selecting a plurality of known cloud detection algorithms;

the traditional foggy detection algorithm comprises a foggy identification algorithm based on contrast, a dark channel first-aid algorithm and the like, wherein an algorithm with higher accuracy rate is selected during selection, and simultaneously, the selected algorithms with different principles are balanced, so that the excessive algorithm occupation of a single principle is avoided;

step S102, preparing a marked group fog picture test sample set;

the cluster fog picture test samples in the cluster fog picture test sample set are the key for obtaining the identification accuracy of each algorithm, and preferably cluster fog pictures which are verified with the data of the visibility detector are adopted;

step S103, testing a plurality of known cluster fog detection algorithms by using the marked cluster fog picture test sample set to obtain a corresponding accuracy matrix A;

the important disadvantage of manual visibility marking of the highway foggy pictures is that objective standards are lacked, the visibility judgment of each person is very subjective, and the marking results of the same marking person for two times of marking the same picture are inconsistent; although the traditional foggy detection algorithm is adopted, the identification precision is required to be provided, if the identification results tend to be consistent for many times, the uncertainty of the labeling can be eliminated;

assuming that n algorithms are provided and 5 visibility levels are provided, the identification accuracy matrix A obtained by testing is as follows:

wherein, the ith group fog recognition algorithm is used for the jth group fogVisibility level identification accuracy A_ijNamely the voting weight of the ith cloud identification algorithm to the jth visibility level;

in the step, firstly, preparing a group fog picture to be annotated to form a group fog picture set to be annotated; then, identifying the to-be-labeled group fog pictures by using the plurality of known group fog detection algorithms to obtain an identification result matrix P; for example:

step S3, obtaining a voting result matrix F by calculating according to the formula F ═ a × P; for example, the voting result matrix F is obtained as follows:

the vote value for the jth visibility level is:

step S4, calculating a maximum vote value in the voting result matrix; for example, in the voting result matrix, the maximum voting value is F₂＝0.65+0.56＝1.21；

in the step, if a plurality of conditions that the maximum voting value is equal to or smaller than a set threshold exist, giving up marking;

The existing traditional visibility detection algorithm, namely the group fog detection algorithm, is mature, but the identification accuracy of visibility grades is not very high, and the visibility grade with the most votes is selected as a marking result by voting through a plurality of group fog detection algorithms, so that scientific automatic marking is realized.

Finally, it should be noted that the above embodiments are only for illustrating the technical solutions of the present invention and not for limiting, and although the present invention has been described in detail with reference to examples, it should be understood by those skilled in the art that modifications or equivalent substitutions may be made on the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, which should be covered by the claims of the present invention.

Claims

1. A highway foggy image visibility grade automatic labeling method comprises the following steps:

step S3, obtaining a voting result matrix F by using the formula F = a × P;

step S4, calculating a maximum vote value in the voting result matrix;

step S6, repeating the steps S2-S5, and automatically marking the visibility grade of all the group fog pictures to be marked in the group fog pictures to be marked;

in step S5, if there are a plurality of equal maximum vote values or the maximum vote values are smaller than the set threshold, discarding the labeling;

step S1 specifically includes:

step S102, preparing a marked group fog picture test sample set;

2. The automatic labeling method for the visibility grade of the highway foggy pictures as claimed in claim 1,

the visibility grade is divided into five grades;

(1) visibility first, representing visibility below 50 meters;

(2) visibility level two, representing visibility 50 to 100 meters;

(3) visibility three-level, representing visibility 100 to 200 meters;

(4) visibility four, representing visibility 200 to 500 meters;

(5) visibility grade five, representing visibility 500 to 1000 meters.