WO2013013487A1

WO2013013487A1 - Device and method for monitoring driving behaviors of driver based on video detection

Info

Publication number: WO2013013487A1
Application number: PCT/CN2011/084857
Authority: WO
Inventors: 徐建闽; 沈文超
Original assignee: 华南理工大学
Priority date: 2011-07-26
Filing date: 2011-12-28
Publication date: 2013-01-31
Also published as: CN102263937A; CN102263937B

Abstract

A device and a method for monitoring driving behaviors of a driver based on video detection. The monitoring device comprises an image capturing and preprocessing module (2), a functional key module (3), an alarm (4), a display (5), an external memory card (6), a power supply module (1) and a central processing unit (7). The monitoring method comprises: performing a device self-test and the central processing unit (7) configuring an internal register of the image capturing and preprocessing module (2); the image capturing and preprocessing module (2) capturing image data and preprocessing an analog image signal; the central processing unit (7) converting a format of the preprocessed data; recognizing a driving behavior of a driver based on video detection; and if it is determined that the behavior is a violation behavior, the alarm (4) sending a corresponding alarm, and storing a corresponding violation image in the external memory card (6) for follow-up verification and tracking. The present invention can effectively monitor whether a driver has a violation behavior during driving, and send an alarm in response to the violation behavior, thereby effectively avoiding traffic accidents resulting from violations of the driver.

Description

Driver driving behavior monitoring device and monitoring method based on video detection

技术领域Technical field

The invention relates to the technical field of automobile safety, and particularly relates to a driver driving behavior monitoring device and a monitoring method based on video detection.

背景技术Background technique

In the driving system consisting of people, cars and roads, the driver is the biggest cause of traffic accidents. In 2009, there were nearly 240,000 road traffic accidents in China and nearly 68,000 deaths. Among them, most of the accidents are caused by driver's operation errors and fatigue driving. Due to changes in age, physical or mental health, mood, etc., even a good driver may not be able to maintain his or her good driving condition for a long time, but the driver himself is hard to realize the gradual attenuation or Regressed. Therefore, monitoring the driver's driving behavior and alerting the violations, improving the driver's driving ability and reducing the driving load, and coordinating the relationship between the driver and the vehicle and the traffic environment, thereby substantially reducing the traffic accident situation. Occurs, it is of great significance.

At present, some research results have been obtained in monitoring driver's driving behavior at home and abroad, which can be roughly divided into two types: one is to judge whether to drink alcohol according to the alcohol content in the driver's exhalation; according to the driver's eyelid and eyeball The relative reflection principle is used to judge whether the driver is fatigued or not; and the driver's driving state is evaluated by judging whether the driver is fatigued or the like according to the driver's brain wave or electrocardiogram to monitor whether the driver is physiologically normal. The other is characterized by the driver's head activity, facial features (such as eyes, head, face) changes, using computer image processing and pattern recognition technology to analyze the driver's driving behavior and mental state. However, these research results are indirectly monitoring and monitoring the driving behavior, and do not directly study the driver's driving behavior itself, there are limitations such as measurement error and high hardware cost.

发明内容Summary of the invention

In order to solve the above problems in the prior art, the present invention provides a driver driving behavior monitoring device and a monitoring method based on video detection. The present invention collects an image including a driver's hand and a steering wheel through a camera, and then processes the image and The identification is to determine whether the driver has an irregularity during driving, and an alert is given to the driver according to the corresponding violation action. The invention can effectively avoid traffic accidents caused by the driver's illegal operation.

The invention is achieved by the following technical solutions:

The invention relates to a driver driving behavior monitoring device based on video detection, comprising: a power module, an image acquisition and preprocessing module, a function button module, an alarm, a display screen, an external memory card and a central processing unit. Wherein: one output end of the power module is connected to the power input end of the image acquisition and preprocessing module, and the other output end of the power module is connected to the power input end of the central processor; the image output interface and the image of the image acquisition and preprocessing module The image input interface of the processor is connected, and the bus interface of the image acquisition and preprocessing module is connected with the bus interface of the central processor; the output end of the function button module is connected with the universal input and output interface of the central processor; the input end of the alarm is The PWM output interface of the central processing unit is connected; the input end of the display is connected to the display output interface of the central processing unit; the external memory card is connected through the central processing unit of the external memory card slot.

In the above-mentioned video detection-based driver driving behavior monitoring device, the power module includes a filter circuit, a transformer circuit, and a backup power source; the power module respectively supplies a working voltage to the camera and the central processing unit; in addition, when the working voltage is When in an abnormal state (below the normal voltage or power off), the backup power supply provides the image acquisition and pre-processing module and the central processing unit with a working voltage for a period of time to prevent data loss caused by the abnormal voltage.

In the above-mentioned video detection-based driver driving behavior monitoring device, the image acquisition and pre-processing module includes a camera and a video processing chip, and an output interface of the camera and an input end of the video processing chip are connected by a video cable, and the central processing The bus interface of the device is connected to the bus interface of the video processing chip in the image acquisition and preprocessing module; the driver's hand and the steering wheel are located in the field of view of the camera, so as to observe the hand motion of the driver; the central processor passes The bus interface configures the internal registers of the video processing chip, so that the image acquisition and pre-processing module has a pre-processing function for the input analog signal, and the pre-processing includes: control of chromaticity and brightness, output data format, and selection of output image synchronization signals. Control, etc.; the preprocessed data is then transmitted to the central processor through the image output interface of the video processing chip.

In the above-mentioned video detection-based driver driving behavior monitoring device, the function button module can be used for manually establishing an area of interest and automatically establishing an area of interest confirmation.

In the above-mentioned video detection-based driver driving behavior monitoring device, the alarm device is mainly used to generate a corresponding alarm when the driver has different illegal driving behavior, and can also be used for device self-checking. .

In the above-mentioned video detection-based driver driving behavior monitoring device, the display screen is mainly used to display the preview processed image.

In the above-mentioned video detection-based driver driving behavior monitoring device, the central processing unit includes: an image input interface, a general-purpose input/output interface, a PWM output interface, a display output interface, an external storage area slot, and a power input. Interface and bus interface. Wherein: the image input interface of the central processing unit is connected with the image output interface of the image acquisition and preprocessing module, and the bus interface of the central processing unit is connected with the bus interface of the image acquisition and preprocessing module, and the universal input and output interface and function of the central processing unit are The output of the button module is connected, the PWM output interface of the central processing unit is connected to the output end of the alarm, the display output interface of the central processing unit is connected to the input end of the display screen, and the central processing unit is connected to the external memory card through the external memory card slot. The power input interface of the central processing unit is connected to the power module. The central processing unit is mainly responsible for the conversion of the image data format, the recognition of the driver's driving behavior based on the video detection, the driving of the function button module and the alarm, the transmission of the data to the display screen and the saving of the data information to the external memory card.

The above-mentioned central processor identifies the driver's driving behavior based on video detection, including: reading the formatted image data, positioning the steering wheel, establishing a region of interest, and extracting hand features (characterizing the region of interest of the original image) Extracting, extracting the hand features (0-1 feature matrix) according to the positional relationship between the driver's hand and the steering wheel, classifying and identifying the library of violation rules, and judging whether the driver's operation is illegal according to the rule base.

The above-mentioned video detection-based driver driving behavior monitoring device has the following monitoring methods: after the power is started, the device self-test and the image acquisition and pre-processing module configure an internal register through the bus interface of the central processor, thereby having an analog input. The function of the signal pre-processing. The camera in the image acquisition and preprocessing module is responsible for acquiring image data, and the video processing chip in the image acquisition and preprocessing module preprocesses the image analog signal, and the preprocessing includes: Chroma and brightness control, Output data format and selection control of output image synchronization signal; preprocessed data is converted by central processor format, and then a driver detection behavior recognition based on video detection is performed. If it is determined to be a violation, the alarm generates a corresponding alarm. And save the corresponding violation picture to the external memory card, and write the violation record to the record file, so as to check the tracking later. The contents of the violation record file include: violation time, violation action, duration, and the number corresponding to the violation image.

The above-mentioned video detection-based driver driving behavior recognition method, the specific steps are as follows:

Step 1: reading the formatted image data, and the read valid image should include information such as the steering wheel and the driver's hand posture (hereinafter referred to as the original image);

Step 2. Position the steering wheel. The read image data is subjected to secondary processing, including grayscale transformation, image filtering, edge extraction, and contour enhancement to obtain an edge image. After secondary processing, the image noise can be eliminated, the detectability of the steering wheel can be enhanced, and the reliability of feature extraction and image recognition can be improved. For the edge image obtained by the secondary processing, the contour of the steering wheel can be extracted and detected by the ellipse fitting algorithm. And positioning;

Step 3. Establish a region of interest. The region of interest is established in the original image according to the positioned steering wheel, and the region of interest includes an area of the steering wheel and the driver's hand information;

In step 4, feature extraction is performed on the region of interest of the original image, mainly hand feature extraction. In this step, the region of interest is divided into M*N sub-regions, and the skin color model is used to determine whether each sub-region contains hand features, and a 0-1 feature matrix is established.

In step 5, the extracted 0-1 feature matrix (representing the hand feature) is classified and recognized according to the positional relationship between the driver's hand and the steering wheel, and the driver's driving action is divided into normal driving, two-hand pressing, two hands off, and the right hand. Off the plate, left hand off the plate, hands crossed on the right hand and hands crossed on the left hand.

Step 6, establishing a violation rule base based on the duration and frequency of the driver's driving action, and determining whether the driver operation is in violation according to the rule base.

In step 2, the shape of the steering wheel in the image is mostly elliptical or circular, and it is necessary to perform ellipse detection on the image to position the steering wheel. Since the steering wheel contour in the car video image has the largest circular or elliptical contour, the edge image obtained by the secondary processing can be extracted by the direct least squares ellipse fitting algorithm and the largest elliptical shape can be detected. To complete the positioning of the steering wheel area.

In step 3, after the positioning of the steering wheel area is completed, the corresponding region of interest is intercepted around the steering wheel. The interception of the region of interest is mainly determined by the positional relationship between the center position and size of the steering wheel and the contour features of the driver's head. At the time of specific interception, the interception is performed by extending the steering wheel region outward by setting the scale factor. For different models, the scale factor is determined experimentally or empirically.

In step 4, the step divides the region of interest into M*N sub-regions, and uses the skin color model to determine whether each sub-region contains hand features, and establishes a 0-1 feature matrix. The simple Gaussian model is selected as the skin color model, and the hand skin color of each sub-area is identified to establish a 0-1 feature matrix, where 1 represents the skin color pixel value and 0 represents the background pixel value. This method is divided into two steps. Firstly, the appropriate skin color model is selected and the parameters of the model are determined. The parameter determination process is as follows: a large number of pixels corresponding to the skin color characteristics are selected as samples, and the distribution is calculated and a skin color Gaussian model is established. The model is then used to determine if the new pixel or region is skin tone. Firstly, select a large number of pixels under normal illumination, strong light, and night (weak light) that meet the characteristics of human skin color under different illuminations, and calculate their distribution and establish a Gaussian model of skin color; then use the model under different illumination to identify new pixels or Whether the area is skin color; which determines the lighting condition according to the image gray distribution.

In step 5, after obtaining the 0-1 feature matrix (representing the hand features), the neural network and the Bayesian network classifier are used to classify and identify them. The specific implementation process is as follows: 1) According to different models, according to a certain proportion And the number of pictures to select various driving behaviors; 2) using the foregoing method to process the selected images to obtain the motion state parameters corresponding to various driving behaviors, to form a training instance set; 3) using the training instance set to neural network and Bayeux The network classifier is trained to obtain a driving behavior classifier adapted to various types of vehicles. Through the classifier, the basic actions of different hands can be identified, a basic illegal driving behavior characteristic database is established, and a basic illegal driving behavior characteristic library is established according to these basic illegal actions.

In step 6, the violation rule is determined by one or more basic violation actions, and the duration of the action, the frequency of occurrence to jointly determine the violation rule; combined with the zone of interest and the rule base of violation rules to determine whether the behavior is in violation.

If the result is a violation according to the above method, the alarm generates a corresponding alarm, and saves the corresponding violation picture to the external memory card, and writes the violation record to the record file, so as to check the tracking later. The contents of the violation record file include: violation time, violation action, duration, and the number corresponding to the violation image.

Since the above scheme is adopted, the present invention has the following advantages and effects:

1. The invention adopts tracking the driver's hand movement to judge whether the driver violates the regulations, and opens up a new direct and effective monitoring path, which is of great significance for preventing traffic accidents caused by illegal driving behavior.

2. The present invention extracts hand features by the central processor for skin detection, selects a simple Gaussian model as the skin color model, takes into account the distribution of pixel points falling within the skin color model, and uses the probability density formula to determine the probability that the pixel points belong to the skin color. Instead of directly classifying all the pixels that fall within the model range into skin color points, the skin color distribution can be better represented than the area model, and the skin color detection efficiency is also much higher, and the parameters of the model are also easy to calculate. Therefore, the device has the advantages of high detection accuracy and high reliability.

3. The invention recognizes the driving behavior of the driver through the central processor. When the driver has an illegal driving operation, a corresponding alarm is generated to remind the driver, and the illegal picture is saved to the external memory card, so as to track the query later, there is a picture. According to evidence, it can effectively reduce the driver's bad driving behavior.

4. The device in the invention has high intelligence, small volume and strong anti-interference, and is convenient for application and promotion.

附图说明 DRAWINGS

FIG. 1 is a schematic structural diagram of a driver's driving behavior monitoring device based on video detection.

2 is a flow chart of a method for identifying driver's illegal driving behavior based on video detection.

具体实施方式 detailed description

The specific implementation of the present invention will be further described below with reference to the accompanying drawings, but the scope of the present invention is not limited thereto.

In the present embodiment, the camera collects an image including the driver's hand and the steering wheel, and determines whether the driver has an irregularity during driving by processing and recognizing the image, and alerts the driver according to the corresponding violation action. The invention can effectively avoid traffic accidents caused by the driver's illegal operation. As shown in FIG. 1 , a driver detection behavior monitoring device based on video detection includes: a power module 1, an image acquisition and preprocessing module 2, a function button module 3, an alarm 4, a display screen 5, and an external memory card 6 And the central processing unit 7. Wherein: one output end of the power module 1 is connected to the power input end of the image acquisition and preprocessing module 2, and the other output end of the power module 1 is connected to the power input end of the central processor 7; the image acquisition and preprocessing module 2 The image output interface is connected to the image input interface of the central processing unit 7, and the bus interface (I ² C bus interface) of the video processing chip in the image acquisition and preprocessing module 2 and the bus interface of the central processing unit 7 (I ² C bus) The interface is connected; the output of the function button module 3 is connected to the universal input/output interface of the central processor 7; the input of the alarm 4 is connected to the PWM output interface of the central processor 7; the input of the display 5 is connected to the central processor The display output interface of 7 is connected; the external memory card 6 is connected through the external memory card slot central processor 7.

The power module 1 includes a filter circuit, a transformer circuit, and a backup power source; the power module 1 provides a working voltage to the camera and the central processing unit respectively; in addition, when the working voltage is in an abnormal state (below the normal voltage or the power is turned off) The backup power supply provides the image acquisition and pre-processing module 2 and the central processing unit 7 with a working voltage for a period of time to prevent data loss caused by the device under abnormal voltage conditions.

The image acquisition and preprocessing module 2 includes a camera and a video processing chip. In the embodiment, the video processing chip selects the SAA7113 chip, and the SAA7113 chip supports input and data output formats of various video signals; the output interface of the camera and the video processing chip the video cable is connected via an input terminal, connected to the central processor 7 I ² C bus interface via the bus interface in the video processing chip (I ² C bus interface) and the preprocessing module 2; driver's steering wheel is in the hand and camera In the field of view, in order to observe the driver's hand movement; the central processor 7 configures the corresponding internal register of the SAA7113 chip through the bus interface (I ² C bus interface), so that the image acquisition and preprocessing module 2 has an input simulation Signal preprocessing function, preprocessing includes: control of chromaticity and brightness, output data format and selection control of output image synchronization signal; image acquisition and preprocessing module 2 works as follows: camera senses environmental change, output PAL The analog signal is transmitted to the image data acquisition and preprocessing module via the video cable. The SAA7113 chip (the SAA7113 chip supports multiple video signal input and data output formats) in the image data acquisition and preprocessing module 2 starts to acquire the PAL analog signal (only for the input one composite video signal), image data acquisition and The video analog output signal of the SAA7113 chip in the pre-processing module 2 is pre-processed and output as a standard 4:2:2 digital standard of the ITU656 protocol, and serves as an image interface interface of the central processing unit 7 (S3C2440). input of.

The function button module 3 can be used to manually establish a region of interest and confirm the automatic establishment of the region of interest.

The alarm 4 is mainly used to generate a corresponding alarm when the driver has different illegal driving, and can also be used for self-checking of the device.

The display screen 5 is mainly used for displaying a preview processed image.

The central processing unit 7, also referred to as an MCU, uses an ARM9 chip of a Samsung S3C2440 microprocessor in this embodiment, with a frequency of 400 MHz and a 133 MHz bus frequency. The central processing unit module 7 includes an image input interface, a general-purpose input/output interface, a PWM output interface, a display output interface, an external memory card slot, a power input interface, and a bus interface (I ² C bus interface). The image input interface of the central processing unit 7 is connected to the image output interface of the image acquisition and preprocessing module 2, and the bus interface of the central processing unit 7 (I ² C bus interface) and the bus interface of the image acquisition and preprocessing module 2 (I ² C bus interface) is connected; the general-purpose input and output interface of the central processing unit 7 is connected to the output end of the function button module 3; the PWM output interface of the central processing unit 7 is connected to the input end of the alarm 4; the display of the central processing unit 7 The output interface is connected to the input end of the display screen 5; the central processing unit 7 is connected to the external memory card 6 of the slot through an external memory card; the power input interface of the central processing unit 7 is connected to the power supply module 1. The central processing unit 7 is mainly responsible for the conversion of the image data format, the recognition of the driver's driving behavior based on the video detection, the driving function button module 3 and the alarm 4, the transmission of the data to the display screen 5, and the saving of the data information to the external memory card.

The above-mentioned central processor identifies the driver's driving behavior based on video detection, including: reading the formatted image data, positioning the steering wheel, establishing a region of interest, and extracting a hand feature (characterizing the region of interest of the original image) Extracting), extracting hand features (0-1 feature matrix) according to the positional relationship between the driver's hand and the steering wheel, classifying and identifying the library of violation rules, and judging whether the driver's operation is illegal according to the rule base.

The above-mentioned video detection-based driver driving behavior monitoring device has the following monitoring method: after the power module 1 is started, the device self-test and the image acquisition and pre-processing module 2 configure internal registers through the bus interface of the central processing unit 7, The image acquisition and pre-processing module 2 thus has a pre-processing function for the input analog signal. The camera in the image acquisition and preprocessing module 2 is responsible for acquiring image data, and the video processing chip in the image acquisition and preprocessing module 2 preprocesses the image analog signal, and the preprocessing includes: Chroma and brightness control, The output data format and the selection control of the output image synchronization signal; the preprocessed data is converted by the central processor 7 image data format, and then a driver detection behavior recognition based on video detection is performed; if it is determined to be a violation, the alarm is generated. Corresponding alarms are saved to the external memory card 6 and the violation records are written to the log file for later verification. The contents of the violation record file include: violation time, violation action, duration, and the number corresponding to the violation image.

The above-mentioned video detection-based driver driving behavior recognition method is shown in FIG. 2, and the specific steps are as follows:

Step 2. Position the steering wheel. The read image data is subjected to secondary processing, including grayscale transformation, image filtering, edge extraction, and contour enhancement, to obtain an edge image. After secondary processing, the image noise can be eliminated, the detectability of the steering wheel can be enhanced, and the reliability of feature extraction and image recognition can be improved. For the edge image obtained by the secondary processing, the contour of the steering wheel can be extracted and detected by the ellipse fitting algorithm. And positioning;

In step 4, feature extraction is performed on the region of interest of the image read in step 1, mainly hand feature extraction. In this step, the region of interest is divided into M*N sub-regions, and the skin color model is used to determine whether each sub-region contains hand features, and a 0-1 feature matrix is established.

In step 2, the shape of the steering wheel in the image is mostly elliptical or circular, and it is necessary to perform ellipse detection on the image to position the steering wheel. Commonly used ellipse detection can be divided into two major methods based on voting and optimization. The representative algorithms of the voting class method include algorithms such as Hough transform and RANSAC. Optimization methods include least squares and genetic algorithms. Due to the large number of elliptic parameters, the focus of voting research is generally on the screening of data points and the use of elliptical geometric properties. Hough transform, RANSAC is a mapping method, projecting sample points into the parameter space, using an accumulator or clustering method to detect the ellipse. This type of algorithm is very robust and can detect multiple ellipses at once, but requires complex operations and a large amount of storage space. Another type of method includes least squares fitting algorithms, genetic algorithms, and other optimized ellipse fitting methods. The main feature of this type of method is its high accuracy, but it cannot be directly used for the detection of multiple ellipses, and it is more sensitive to noise than the former method. Since the steering wheel profile in the car video image has the largest circular or elliptical contour, that is, only one largest elliptical shape needs to be detected, the present embodiment adopts a direct least squares ellipse fitting algorithm to extract the steering wheel profile and detect the largest. The shape of the ellipse is detected, and the circumscribed rectangle of the ellipse is detected to complete the positioning of the steering wheel area.

In step 3, the interception of the region of interest is mainly determined by the positional relationship between the center position and size of the steering wheel and the contour features of the driver's head. In the specific interception, after the positioning of the steering wheel area is completed, the interception is performed by extending from the steering wheel area with a certain proportional coefficient. For different models, the scale factor is calibrated experimentally or can be determined empirically.

In step 4, feature extraction is performed on the intercepted region, mainly a hand feature, and the hand feature can be extracted by detecting the skin. In this step, the region of interest is divided into M*N sub-regions, and the skin color model is used to determine whether each sub-region contains hand features, and a 0-1 feature matrix is established.

In order to detect the skin, it is necessary to select a suitable skin color model to identify the skin color of the hand in the intercepted area. The skin color models commonly used in image processing are roughly divided into two categories: simple threshold segmentation and probability model. Among them, the probability model has a histogram model, a simple Gaussian model and a mixed Gaussian model. The simple Gaussian model is a model that assumes that the skin color distribution is a unimodal Gaussian distribution. The simple Gaussian model takes into account the distribution of pixels falling within the skin model, and uses the probability density formula to determine the probability that a pixel belongs to the skin color, instead of simply classifying all pixels falling within the model into skin color points. The regional model can better represent the distribution of skin color, so its skin color detection efficiency is relatively high, and the parameters of the model are easy to calculate. In the present embodiment, a simple Gaussian model is used as a skin color model to identify the hand skin color of each sub-region. A 0-1 feature matrix is established, where 1 represents the skin color pixel value and 0 represents the background pixel value. This method is divided into two steps. Firstly, the appropriate skin color model is selected and the parameters of the model are determined. The parameter determination process is as follows: a large number of pixels corresponding to the skin color characteristics are selected as samples, and the distribution is calculated and a skin color Gaussian model is established. The model is then used to determine if the new pixel or region is skin tone. Firstly, select a large number of pixels under normal illumination, strong light, and night (weak light) that meet the characteristics of human skin color under different illuminations, and calculate their distribution and establish a Gaussian model of skin color; then use the model under different illumination to identify new pixels or Whether the area is skin color, thus establishing a 0-1 feature matrix. Which light condition is determined according to the image gray distribution.

In step 5, the extracted 0-1 feature matrix (representing hand features) is classified and identified. The method of the invention adopts a neural network and a Bayesian network classifier to classify and identify the 0-1 feature matrix, and identify different feature matrices to distinguish whether it is an illegal driving behavior. After obtaining the 0-1 feature matrix, it is classified and identified by neural network and Bayesian network classifier. The specific implementation process is as follows: 1) According to different vehicle models, select pictures of various driving behaviors according to a certain proportion and quantity; 2) Using the foregoing method to process the selected images to obtain the state parameters corresponding to various driving behaviors, and form a training instance set; 3) training the neural network and the Bayesian network classifier with the training instance set to adapt to various types of vehicles Driving behavior classifier. Through the classifier, different basic movements of the hand can be identified, and a basic illegal driving behavior characteristic library is established. In this embodiment, seven basic actions can be identified by the classifier: normal driving, two-hand pressing, hands off, right hand off, The left hand is off the plate, the hands are crossed, the right hand is on the top, and the hands are crossed on the left hand.

If the result is a violation according to the above method, the alarm generates a corresponding alarm, and saves the corresponding violation picture to the external memory card 6, and writes the violation record into the record file, so as to check the tracking later. The contents of the violation record file include: violation time, violation action, duration, and the number corresponding to the violation image.

The embodiment can automatically recognize whether there is a hand violation during driving, provide an alarm to the driver, and write the violation record into the log file for later check tracking. The accuracy of the embodiment is high, and the occurrence of a traffic accident caused by the driver's illegal operation can be effectively avoided.

Claims

1 The driver detection behavior monitoring device based on video detection is characterized by comprising a power module, an image acquisition and preprocessing module, a function button module, an alarm, a display screen, an external memory card and a central processing unit, wherein: one of the power modules The output end is connected to the power input end of the image acquisition and preprocessing module, and the other output end of the power module is connected to the power input end of the central processing unit; the image output interface of the image acquisition and preprocessing module and the image input interface of the central processing unit Connected, and the bus interface of the image acquisition and preprocessing module is connected to the bus interface of the central processor; the output of the function button module is connected with the universal input and output interface of the central processor; the input of the alarm and the PWM output of the central processor Interface connection; the input end of the display is connected to the display output interface of the central processing unit; the external memory card is connected through the central processing unit of the external memory card slot; the function button module is used for manually establishing the region of interest and automatically establishing an interest Confirmation of the area; the alarm described for driving When irregularities occur in different driving behavior, generating an alarm to alert appropriate, but also a device for self-check prompt; the display screen, image processing for displaying the preview.

2 The monitoring device according to claim 1, wherein the power module comprises a filter circuit, a transformer circuit and a backup power source; the power module respectively supplies a working voltage to the camera and the central processor; and when the working voltage is in an abnormal state The backup power supply provides the image acquisition and pre-processing module and the central processing unit with a working voltage for a period of time.

3 The monitoring device according to claim 1, wherein the image acquisition and preprocessing module comprises a camera and a video processing chip, and an output interface of the camera and an input end of the video processing chip are connected via a video cable, and a bus of the central processing unit The interface is connected to the bus interface of the video processing chip in the image acquisition and preprocessing module; the driver's hand and the steering wheel are located in the field of view of the camera; the central processor configures the internal processing register of the video processing chip through the bus interface, and the image is collected and pre-processed. The processing module has a pre-processing function for inputting an analog signal, and the pre-processing includes: control of chromaticity and brightness, selection of output data format and output image synchronization signal; and pre-processed data is transmitted through an image output interface of the video processing chip Go to the central processor.

4 The monitoring device according to claim 1, wherein the central processing unit comprises an image input interface, a universal input/output interface, a PWM output interface, a display output interface, an external storage area slot, a power input interface, and a bus interface. The image input interface of the central processing unit is connected to the image output interface of the image acquisition and preprocessing module, and the bus interface of the central processing unit is connected with the bus interface of the image acquisition and preprocessing module, and the central processing unit has a general input/output interface and function buttons. The output end of the module is connected, the PWM output interface of the central processing unit is connected to the output end of the alarm, the display output interface of the central processing unit is connected to the input end of the display screen, and the central processing unit is connected to the external memory card through the external memory card slot. The power input interface of the central processing unit is connected to the power supply module; the central processing unit is mainly responsible for converting the image data format, The driver's driving behavior is recognized based on video detection, the function button module and the alarm are driven, the data is transmitted to the display, and the data information is saved to the external memory card.

5 The monitoring device according to claim 1, wherein the identification of the driver's driving behavior based on the video detection by the central processor comprises: reading the formatted image data, positioning the steering wheel, establishing a region of interest, The hand features are extracted, and the extracted hand features are classified and identified according to the positional relationship between the driver's hand and the steering wheel. Violation of the rule base, and based on the rule base to determine whether the driver's operation is in violation.

6 The method for monitoring driving behavior of a driver according to any one of claims 1 to 5, characterized in that after the power is turned on, the device self-test, the image acquisition and pre-processing module configures an internal register through a bus interface of the central processing unit, Therefore, the function of preprocessing the input analog signal is performed; the camera in the image acquisition and preprocessing module is responsible for acquiring image data, and the video processing chip in the image acquisition and preprocessing module preprocesses the image analog signal, and the preprocessing includes: Control of chromaticity and brightness, output data format and selection control of output image synchronization signal; preprocessed data is converted by central processor format, and driver driving behavior recognition based on video detection is performed. If it is judged to be a violation, the alarm generates a corresponding alarm, and saves the corresponding violation picture to the external memory card, and writes the violation record to the record file for later check and tracking; the content of the violation record includes violation time, violation action, Duration and number of the corresponding violation image.

7. The monitoring method according to claim 6, wherein the driver driving behavior recognition based on the video detection comprises the following steps:

Step 1: reading the formatted image data, the read effective image should include the steering wheel and the driver's hand posture information;

Step 2: Positioning the steering wheel, performing secondary processing on the read image data, including four processing steps of gradation transformation, image filtering, edge extraction, and contour enhancement to obtain an edge image; and utilizing the edge image obtained by the secondary processing The ellipse fitting algorithm extracts, detects and locates the contour of the steering wheel;

Step 3: establishing a region of interest, where the original map establishes a region of interest according to the positioned steering wheel, and the region of interest includes an area of the steering wheel and the driver's hand information;

Step 4: Perform feature extraction on the region of interest of the original image, mainly including hand feature extraction. This step divides the region of interest into M*N sub-regions, and uses the skin color model to determine whether each sub-region contains hand features, and establishes 0. -1 feature matrix;

Step 5: classifying and identifying the extracted 0-1 feature matrix according to the positional relationship between the driver's hand and the steering wheel, and distinguishing the driver's driving action into normal driving, Press the plate with both hands, hands off the plate, right hand off the plate, left hand off the plate, hands crossed on the right hand and hands crossed on the left hand;

Step 6, establishing a violation rule base based on the duration and frequency of the driver's driving action, and according to the rule base Determine if the driver's operation is illegal.

8 The monitoring method according to claim 7, wherein in step 2, the steering wheel shape in the image is mostly elliptical or circular, and the least squares ellipse fitting algorithm is used to extract the contour of the steering wheel and detect the maximum. Elliptical shape to complete the positioning of the steering wheel area;

9 The monitoring method according to claim 7, wherein in step 3, after the positioning of the steering wheel area is completed, the corresponding region of interest is intercepted around the steering wheel; the interception of the region of interest is mainly caused by the center position and size of the steering wheel. Corresponding to the positional relationship of the contour features of the driver's head; in the specific interception, the interception is performed from the steering wheel area by setting the scale factor outward; for different models, the proportional coefficient is determined by experiment or empirically;

In step 4, the step divides the region of interest into M*N sub-regions, and uses the skin color model to determine whether each sub-region contains hand features, and establishes a 0-1 feature matrix; and selects a simple Gaussian model as a skin color model for each child. The hand skin color of the area is identified, and a 0-1 feature matrix is established, where 1 represents the skin color pixel value and 0 represents the background pixel value;

In step 5, after obtaining the 0-1 feature matrix, the neural network and the Bayesian network classifier are used to classify and identify the 0-1 feature matrix. Specifically, according to different vehicle models, various driving behavior pictures are selected according to a certain proportion and quantity; The selected steps are used to process the selected images to obtain the motion state parameters corresponding to various driving behaviors, and form a training instance set. The training examples are used to train the neural network and the Bayesian network classifier to obtain driving suitable for various types of vehicles. Behavior classifier; can pass the classifier Identify basic movements of different hands, establish a basic illegal driving behavior characteristic database, and then establish a basic illegal driving behavior characteristic library according to these basic illegal actions;

In step 6, the violation rule jointly determines the violation rule by more than one basic violation action, the duration of the violation action, and the frequency of occurrence; Combine the area of interest and the rule base of violation rules to determine whether the behavior is in violation; If the judgment result is a violation, the alarm generates a corresponding alarm, saves the corresponding violation picture to the external memory card, and writes the violation record to the record file, so as to check the tracking later.

10, The monitoring method according to claim 7, wherein in step 4, pixel points conforming to the skin color feature of the human body are selected as samples, the distribution thereof is calculated and a skin color Gaussian model is established; and then the model is used to determine whether the new pixel or region is Skin color; firstly select normal light, glare, and a large number of pixels in the night to meet the characteristics of human skin color under different illuminations, count the distribution and establish a skin color Gaussian model; then use the model under different illumination to determine whether the new pixel or region is It is the skin color; it determines which lighting condition is based on the image gray distribution.