CN109800686A

CN109800686A - A kind of driver's smoking detection method based on active infrared image

Info

Publication number: CN109800686A
Application number: CN201811648701.0A
Authority: CN
Inventors: 仇开金; 许端; 王述良; 程建伟
Original assignee: Wuhan Jimu Intelligent Technology Co Ltd
Current assignee: Wuhan Jimu Intelligent Technology Co Ltd
Priority date: 2018-12-30
Filing date: 2018-12-30
Publication date: 2019-05-24

Abstract

Driver's smoking detection method based on active infrared image that the present invention provides a kind of, based on vehicle-mounted active infrared image, driver's face, locating human face's key point (eyes, nose, mouth), the outer profile for judging cigarette are detected using image steganalysis method, cigarette is confirmed using deep learning method, final realize judges the function whether driver smokes.Operation in whole day of the present invention 24 hours, Detection accuracy is high, computation complexity is low.

Description

A kind of driver's smoking detection method based on active infrared image

Technical field

The invention belongs to image identification technical fields, and in particular to a kind of driver's smoking inspection based on active infrared image Survey method.

Background technique

With the rapid development of the automotive industry, car owning amount increases rapidly, to safe driving propose new demand and Challenge.One of the main reason for cigarette smoking in driving procedure is initiation traffic accident, existing smoking detection method is main There are sensor-based smog detection method and the smoking detection method based on image.

Common smoke sensor device has: ion type smog sensor, inside are sensed using ion type smog, are widely used in In each fire protection alarm system；Photoelectric smoke sensor inside has optics labyrinth and infrared tube, and infrared receiving tube is received not when smokeless The infrared light issued to infrared transmitting tube, when flue dust enters optics labyrinth, by reflecting and reflecting, reception pipe receives infrared Light, intelligent alarm circuit judges are if it exceeds threshold value issues alarm；Gas-sensitive smog sensor mainly includes Semiconductor gas sensors Sensor, catalytic combustion type gas sensor and Electro-chemical Gas Sensor etc., most commonly used is semiconductor gas sensor. Driver, which opens a window, to divulge information and reduces smokescope, and will lead to smoke sensor device can't detect smog and missing inspection occur.

The existing smoking detection method based on image is based on visible images, and with computer vision technique, judgement is driven Whether the person of sailing smokes, and different computer vision methods often obtain completely different performance and effect, currently used Method includes traditional mode identification method and deep learning method.When night is rather dark, visible light is ineffective, and counts Traditional mode recognition methods in calculation machine visible sensation method leads to missing inspection or erroneous detection often due to detection accuracy is low, and the practicability is poor；And Depth learning technology in computer vision methods causes hardware cost high, is unfavorable for pushing away on a large scale since computation complexity is big Wide application.

Summary of the invention

The technical problem to be solved by the present invention is providing a kind of driver's smoking detection side based on active infrared image Method is based on infrared image, judges whether driver smokes using computer vision technique.

A kind of technical solution taken by the invention to solve the above technical problem are as follows: 1. driving based on active infrared image The person's of sailing smoking detection method, comprising the following steps:

Step S1: the active infrared image of vehicle carried driving position is obtained.

Step S2: with image bilinearity difference, histogram equalization and median filter respectively to obtaining in step S1 Image is zoomed in and out and is filtered, and exports treated image.

Step S3: to the image obtained in step S2 with image characteristics extraction device extract feature, with classifier to feature into Row classification, the position for obtaining meeting classifier is face location；In magnitude order by face, obtaining maximum face is Driver's face.

Step S4: face key point is detected to the image obtained in step S3 detection algorithm, exports the key point of face Position.

Step S5: parallel lines detection is carried out to the image obtained in step S4, obtains the candidate region of cigarette.

Step S6: parallel lines filtering is carried out to the image obtained in step S5, filters out the corresponding parallel lines of non-cigarette.

Step S7: detecting the image obtained in step S6, if still having parallel lines, uses deep neural network Classifier classifies to the image-region of existing parallel lines, judges whether it is cigarette, then counts plus one if it is cigarette, such as Fruit be not count it is constant.

Step S8: judging the obtained counting of step S7, if counting is more than threshold value N within time continuous T second, It then alerts, does not otherwise alert.

According to the above scheme, in the step S3, with the HOG of the image obtained in image characteristics extraction device extraction step S2 Feature classifies to HOG feature with SVM classifier, obtains the position for meeting SVM classifier, as face location；By face In magnitude order, maximum face, as driver's face are obtained.

According to the above scheme, in the step S3, special with the local grain of the image obtained in LBP operator extraction step S2 Sign, classifies to Local textural feature with Adaboost algorithm, obtains the position for meeting Adaboost algorithm requirement, that is, is people Face position；In magnitude order by face, maximum face, as driver's face are obtained.

According to the above scheme, in the step S4, to the image obtained in step S3 SDM algorithm or ASM algorithm or AAM Algorithm or CLM algorithm or SRA algorithm detect face key point, export the key point position of face.

According to the above scheme, in the step S5, specific steps are as follows:

Step S51: mouth region is obtained with the position of mouth in face key point, and region is expanded, after being expanded Region correspondence image, is denoted as G0.

Step S52: line segment is detected in G0 with LSD line segment detecting method, line segment aggregate is obtained, is denoted as G1；It deletes in G1 Length is less than the line segment of threshold value L0, and the line-segment sets after deletion are denoted as G2.

Step S53: all angles are searched in G2 greater than threshold value A 0 and the line segment pair of similar length, are denoted as G3, then in G3 Line segment to the candidate region of as cigarette.

According to the above scheme, in the step S6, specific steps are as follows:

Step S61: distance is searched in G3 and is greater than threshold value D0 and is less than the line segment pair of threshold value D1, is denoted as G4.

Step S62: cannot will meet simultaneously in G4 an endpoint at a distance from mouth by-level line be less than threshold value D3 and For line segment of the image brightness values less than threshold value D4 to filtering out, output meets the line segment pair of condition at another endpoint, is denoted as G5.

The invention has the benefit that

The detection method 1. a kind of driver based on active infrared image of the invention smokes, realizes and is based on infrared image, utilize Computer vision technique judges the function whether driver smokes.

2. the present invention uses active infrared image, have the advantages that run for whole day 24 hours.

3. missing inspection of the present invention and erroneous detection are low, Detection accuracy is high.

4. calculation amount of the present invention is small, low to hardware computing resource demand, promoted convenient for a wide range of.

Detailed description of the invention

Fig. 1 is the flow chart of the embodiment of the present invention.

Specific embodiment

The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

Referring to Fig. 1, detection method includes the following steps for a kind of driver's smoking based on active infrared image of the invention:

Step S2: with image bilinearity difference, histogram equalization and median filter respectively to obtaining in step S1 Image is zoomed in and out and is filtered, and for balancing picture contrast, removal noise, and exports treated image.

Step S3: to the image obtained in step S2 with image characteristics extraction device extract feature, with classifier to feature into Row classification, the position for obtaining meeting classifier is face location；In magnitude order by face, obtaining maximum face is Driver's face；Wherein HOG feature (Histogram of Oriented Gradient) can be selected in image characteristics extraction algorithm Extraction or LBP operator (Local Binary Pattern)；SVM classifier (Support Vector can be selected in classifier ) or Adaboost algorithm Machine.

Step S4: face key point is detected to the image obtained in step S3 detection algorithm, exports the key point of face Position；Wherein SDM (Supervised Descent Method) algorithm, ASM (Active Shape can be selected in detection algorithm Model) algorithm, AAM (Active Appearance Model) algorithm, CLM (Constrained Local Model) algorithm, SRA (shape regression approach) algorithm；The key point of face includes the positions such as eyes, nose, mouth.

Step S5: parallel lines detection is carried out to the image obtained in step S4, obtains the candidate region of cigarette, specific steps Are as follows:

Step S52: line segment is detected in G0 with LSD line segment detecting method (Line Segment Detector), obtains line Duan Jihe is denoted as G1, deletes the line segment that length in G1 is less than threshold value L0, the line-segment sets after deletion are denoted as G2.

Step S6: parallel lines filtering is carried out to the image obtained in step S5, filters out the corresponding parallel lines of non-cigarette, is had Body step are as follows:

The detection method in conclusion a kind of driver based on active infrared image of the invention smokes, is based on vehicle-mounted master Dynamic infrared image, using traditional image steganalysis method detection driver's face, locating human face's key point (eyes, nose, Mouth), judge the outer profile of cigarette, cigarette is confirmed using the method for deep learning, final realize judges that driver is The function of no smoking.The present invention uses active infrared image, has the advantages that run for whole day 24 hours；Missing inspection of the present invention and erroneous detection Low, Detection accuracy is high；Calculation amount of the present invention is small, low to hardware computing resource demand, promotes convenient for a wide range of.

Above embodiments are merely to illustrate design philosophy and feature of the invention, and its object is to make technology in the art Personnel can understand the content of the present invention and implement it accordingly, and protection scope of the present invention is not limited to the above embodiments.So it is all according to It is within the scope of the present invention according to equivalent variations made by disclosed principle, mentality of designing or modification.

Claims

The detection method 1. a kind of driver based on active infrared image smokes, it is characterised in that: the following steps are included:

Step S1: the active infrared image of vehicle carried driving position is obtained；

Step S2: with image bilinearity difference, histogram equalization and median filter respectively to the image obtained in step S1 It zooms in and out and is filtered, and export treated image；

Step S3: feature is extracted to the image obtained in step S2 image characteristics extraction device, feature is divided with classifier Class, the position for obtaining meeting classifier is face location；In magnitude order by face, obtaining maximum face is to drive Member's face；

Step S4: face key point is detected to the image obtained in step S3 detection algorithm, exports the key point position of face；

Step S5: parallel lines detection is carried out to the image obtained in step S4, obtains the candidate region of cigarette；

Step S6: parallel lines filtering is carried out to the image obtained in step S5, filters out the corresponding parallel lines of non-cigarette；

Step S7: detecting the image obtained in step S6, if still having parallel lines, is classified with deep neural network Device classifies to the image-region of existing parallel lines, judges whether it is cigarette, then counts if it is cigarette and adds one, if not Be count it is constant；

Step S8: judging the obtained counting of step S7, if counting is more than threshold value N within time continuous T second, then accuses It is alert, otherwise do not alert.
The detection method 2. driver according to claim 1 based on a kind of based on active infrared image smokes, feature It is: in the step S3, with the HOG feature of the image obtained in image characteristics extraction device extraction step S2, uses svm classifier Device classifies to HOG feature, obtains the position for meeting SVM classifier, as face location；In magnitude order by face, Obtain maximum face, as driver's face.
The detection method 3. driver according to claim 1 based on a kind of based on active infrared image smokes, feature It is: in the step S3, with the Local textural feature of the image obtained in LBP operator extraction step S2, uses Adaboost Algorithm classifies to Local textural feature, obtains the position for meeting Adaboost algorithm requirement, as face location；By face In magnitude order, maximum face, as driver's face are obtained.
The detection method 4. driver according to claim 1 based on a kind of based on active infrared image smokes, feature It is: in the step S4, to the image obtained in step S3 SDM algorithm or ASM algorithm or AAM algorithm or CLM algorithm Or SRA algorithm detects face key point, exports the key point position of face.
The detection method 5. driver according to claim 1 based on a kind of based on active infrared image smokes, feature It is: in the step S5, specific steps are as follows:

Step S51: mouth region is obtained with the position of mouth in face key point, and region is expanded, after being expanded Region correspondence image, is denoted as G0；

Step S52: line segment is detected in G0 with LSD line segment detecting method, line segment aggregate is obtained, is denoted as G1；Delete length in G1 Less than the line segment of threshold value L0, the line-segment sets after deletion are denoted as G2；

Step S53: all angles are searched in G2 greater than threshold value A 0 and the line segment pair of similar length, are denoted as G3, then the line in G3 Candidate region of the section to as cigarette.
The detection method 6. driver according to claim 1 based on a kind of based on active infrared image smokes, feature It is: in the step S6, specific steps are as follows:

Step S61: distance is searched in G3 and is greater than threshold value D0 and is less than the line segment pair of threshold value D1, is denoted as G4；

Step S62: it cannot will meet simultaneously an endpoint in G4 and be less than threshold value D3 and another at a distance from mouth by-level line For line segment of the image brightness values less than threshold value D4 to filtering out, output meets the line segment pair of condition at a endpoint, is denoted as G5.