CN102360499B - Multi-lane line tracking method based on Kalman filter bank - Google Patents

Abstract

The invention discloses a multi-lane line tracking method based on a Kalman filter bank, which belongs to the technical field of digital image processing. By establishing a Kalman filter model for lane line tracking, the Kalman filter model is used to output the optimal estimated value of the lane line on the basis of lane line detection on the collected road images in front of the vehicle; when 25 consecutive road images When the same lane line is detected, the optimal estimate of the Kalman filter is considered valid, and the optimal estimate output by the Kalman filter is taken as the detection result. The invention makes the detection result of the lane line in the lane departure warning system more stable, thereby greatly reducing the inaccurate or missed detection of the lane line caused by the shaking of the vehicle body, the change of illumination, the damage of the road surface, etc.; The corresponding number of lane lines can be tracked at the same time, which greatly improves the accuracy of lane line detection and anti-interference ability compared with the single tracking of the area of interest of the lane line.

Description

A multi-lane tracking method based on Kalman filter bank

technical field

The invention belongs to the technical field of digital image processing, and mainly relates to a method for simultaneously tracking multiple lane lines detected in road video images. the

Background technique

Lane Departure Warning System (Lane Departure Warning System, LDWS) is a system that assists the driver to reduce traffic accidents caused by lane departure by means of an alarm. The continuous progress of social economy and the rapid development of modern computer technology, the integration of new technologies such as multi-sensor technology and communication control theory, make the popularization and application of LDWS possible, thereby maximizing the safety of driving and reducing social and economic losses. reduced to a minimum. the

At present, the implementation of LDWS is based on the method of computer vision. First, the lane line is recognized in the collected road video images, and then it is judged whether to issue a departure warning according to the relationship between the vehicle position and the lane line. Because the lane line is a reference for judging whether the vehicle has deviated, the first condition for the realization of LDWS is to accurately detect the lane line through digital image processing. However, during the driving process of the vehicle, due to vehicle body shaking, light changes, road pollution, and lane line damage, etc., the detection results of the same lane line in the collected road video images will be unstable, resulting in a certain frame or several frames. If the image completely deviates from or loses the target, it will affect the inaccuracy of LDWS early warning. the

Therefore, after the lane lines are detected by digital image processing, the target tracking method must be used to use the detected lane lines as samples for sample training to predict the optimal estimated value and complete the lane line tracking. The final output of the optimal estimated value replaces the observed value for early warning, which will make the working state of LDWS more stable and greatly reduce the impact of lane line detection errors on LDWS performance. Therefore, the present invention proposes to use a Kalman filter group to track multiple detected lane lines at the same time, which can greatly improve the detection accuracy of lane lines. the

At present, there are several ways to realize the target tracking of lane lines:

1. Input the coordinates of the quadrilateral area completely including the detected lane line into the Kalman filter as sample information, so as to predict the area containing the lane line in the next sequence image, and define it as the region of interest (Region of Interest, ROI), in order to speed up the speed and accuracy of lane line detection, a single Kalman filter is used to track the region of interest containing lane lines. Although this method speeds up the detection speed, when the lane line cannot be detected in the prediction result ROI, the system cannot work normally. This method is detailed in the literature: Zheng Banggui, Tian Bingxiang, Duan Jianmin. Lane Recognition Technology Based on Kalman Prediction and Back Projection. Computer Engineering and Design. 2009,30(6). 

2. Prediction of the small window of the region of interest. According to the lane lines that have been detected, define a small area that only contains a part of a lane line, and input the coordinates of this small area as sample information into the Kalman filter to predict the next small area that contains this lane line. This method is for 1 The improvement of the method narrows the area of interest from including all lanes to only a certain part of a lane, which improves the detection accuracy of lanes, but still cannot completely avoid the problem that LDWS cannot detect lanes normally. This method is detailed See literature: Yu Houyun, Zhang Weigong. Lane Line Recognition and Tracking Based on Dynamic Region of Interest. Industrial Instrumentation and Automation Devices. Issue 5, 2009.

3. Use other filter algorithms to track lane lines, such as using extended Kalman filter technology and image processing technology to establish a dynamic visual window for lane tracking, and then extract the lane boundaries. For details, see the literature: Chen Ying, Han Chongzhao. Lane Fusion Tracking Based on Extended Kalman Filter. Highway Traffic Science and Technology. Vol.12, No.12, Nov.2004. There is also an effective ROI screening algorithm for lane markings based on the combination of contrast and Otsu Law , and carried out adaptive threshold image segmentation on the effective ROI. Through the effective screening of the ROI, the influence of the discontinuity or damage of the lane markings can be reduced. For details, see the literature: Guan Xin, Jia Xin, Gao Zhenhai. Interested in lane detection Region Selection and Adaptive Threshold Segmentation. Highway Traffic Science and Technology. Vol.26, No.6, Jun.2009.

The above-mentioned tracking algorithms use a single filter to track the lane line ROI, and then identify the lane line in the ROI to indirectly complete the lane line tracking. With the continuous development of the economy and society, the lanes of high-grade roads are getting wider and wider, and the imaging quality of image acquisition equipment is getting higher and higher, which will inevitably lead to multiple lane lines in one image. At this time, it is difficult to use the above tracking algorithm accurately. Each detected lane line is tracked. In this context, the present invention uses a Kalman filter bank to directly track multiple detected lane line objects without tracking the lane line ROI area. the

Contents of the invention

The present invention proposes a multi-lane line tracking method based on a Kalman filter group. The method does not need to pre-set the area of interest of the lane line, and can directly track the detected multi-lane lines. This method can greatly improve the accuracy of lane line detection without affecting the real-time requirements of LDWS. the

In order to describe the content of the present invention conveniently, some terms are defined first. the

Definition 1: Kalman filtering. Kalman Filter (KF) is a recursive data processing algorithm for estimating the state of a linear dynamic system. Through the dynamic update of sample training, it continuously predicts and describes the next state and measurement data. The principle and implementation process of the Kalman filter are detailed in the literature: He Zishu. Modern digital signal processing and its application. Tsinghua University Press. Chapter 7: Kalman filter. 

Definition 2: Hough transform. That is, the Hough transform uses the duality between the straight line plane space and the Hough plane points and lines, that is, the collinear points in the image straight line space correspond to the intersecting lines in the Hough plane; conversely, all the Hough planes intersect at the same point Lines have collinear points corresponding to them in the image space. The Hough transform can transform the linear feature search problem into the Hough plane maximum search problem, which is one of the most widely used algorithms in the field of linear feature extraction. For the principle and implementation process, please refer to the literature: Gonzales. Digital Image Processing. Electronic Industry Press. Chapter 10: Image Segmentation. 

A multi-lane line tracking method based on Kalman filter bank, comprising the following steps:

Step 1: During the driving process of the vehicle, the vehicle-mounted video acquisition equipment is used to collect the road image in front of the vehicle; and the collected road image data is input into the computer for lane line detection. the

Use the Hough transform to detect the lane lines, and use the lateral distance ρ and the azimuth angle θ to represent the detected lane lines, that is, (ρ _i , θ _i ) (where i=1, 2, 3... represent the number of lane lines), Indicates the lane lines detected on the road image. For the detailed algorithm of lane line detection using Hough transform, please refer to the literature: Liu Tan. Research on Vehicle Driving Assistance Technology Based on Monocular Vision. University of Electronic Science and Technology of China Master's Degree Thesis. 2009.09.17.

Step 2: Establish a Kalman filter model for lane tracking. the

Step 2-1: Define the state of the lane line;

Since the motion of the vehicle during the acquisition of two consecutive frames of video images can be regarded as a uniform motion, the state of the lane line detected in each frame of road image is represented by a state vector X(k):

X(k)=[ρ(k), θ(k), u(k), ω(k)] ^T

表示车辆在状态k，即t时刻的径向速度，表示车辆在状态k，即t时刻的角速度。  in

Indicates the radial velocity of the vehicle at state k, that is, at time t, Indicates the angular velocity of the vehicle at state k, that is, at time t.

Step 2-2: Define the state equation of the Kalman filter in the lane tracking process;

Because the system satisfies the following formula from state X(k) to the next state X(k+1):

$\left\{\begin{array}{c}\mathrm{<span\; class="notranslate">\; \&rho;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; \&rho;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>\\ \mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&omega;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>\\ \mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>\\ \mathrm{<span\; class="notranslate">\; \&omega;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; \&omega;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; 4</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>\end{array}\right.$

The system noise of the set state k is: W(k)=[w ₁ (k), w ₂ (k), w ₃ (k), w ₄ (k)] ^T , so the state equation of the Kalman filter expresses for:

X(k)=AX(k-1)+W(k)

in $A=\left[\begin{array}{cccc}1& 0& 1& 0\\ 0& 1& 0& 1\\ 0& 0& 1& 0\\ 0& 0& 0& 1\end{array}\right]$

Step 2-3: Define the observation equation of the lane line tracking process;

Since the observed value is the result of the actual state value and the influence of noise, the observed value can be expressed as follows:

$\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left[\begin{array}{c}{\mathrm{<span\; class="notranslate">\; z</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span>\\ {\mathrm{<span\; class="notranslate">\; z</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span>\end{array}\right]<span\; class="notranslate">\; =</span>\left[\begin{array}{c}\mathrm{<span\; class="notranslate">\; \&rho;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span>\\ \mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span>\end{array}\right]$

So the observation equation is obtained as:

Z(k)=HX(k)+V(k)

where the measurement matrix $h=\left[\begin{array}{cccc}1& 0& 0& 0\\ 0& 1& 0& 0\end{array}\right],$ V(k) is the observation noise, and V(k)=[v ₁ (k), v ₂ (k)] ^T

Step 2-4: Define the Kalman filter iterative update equation:

X(k|k)=X(k|k-1)+Kg(k)(Z(k)-HX(k|k-1)) 

in:

Kg(k)=P(k|k-1)H ^T /(HP(k|k-1)H ^T +R)

P(k|k)=(I-Kg(k)H)P(k|k-1)

X(k|k) is the optimal estimated value of the k state output by the Kalman filter, X(k|k-1) is the result of the k state predicted by the k-1 state, and Kg(k) is the state k The lower Kalman filter gain, Z(k) is the observed value of state k, and H is the measurement matrix. the

Step 3: System initialization

Step 3-1: Set the initial value of the state vector;

According to the basic principle of Kalman filter, as long as the initial state satisfies E[X(0)]=E[X(0|0)], the estimated value obtained is unbiased, so the initial state of the system is set as the zero vector ,Right now

X(0)＝[0 0 0 0] ^T

Step 3-2: Set the initial value of system noise;

The system noise W(k) is a zero-mean random process vector, and the system noise is statistically independent at different time states, and its correlation matrix satisfies the following formula:

$\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; [</span>\mathrm{<span\; class="notranslate">\; W</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; W</span>}}^{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; Q</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; \&delta;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\mathrm{<span\; class="notranslate">\; Q</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>& \mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; no</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>& \mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; \&NotEqual;</span>\mathrm{<span\; class="notranslate">\; no</span>}\end{array}\right.$

因此设定系统噪声初始值为：  It is assumed that the noises of different states at the same time are also statistically independent, so the matrix Q(k) is a diagonal matrix. Because the system noise is caused by the maneuverability of the vehicle, it has little influence in the tracking process. For the convenience of calculation, it is assumed that w ₁ (k), w ₂ (k), w ₃ (k), and w ₄ (k) are independent and distribution, and satisfy

Therefore, the initial value of the system noise is set as:

$\mathrm{<span\; class="notranslate">\; Q</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; [</span>\mathrm{<span\; class="notranslate">\; W</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; W</span>}}^{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span><span\; class="notranslate">\; =</span>\left[\begin{array}{cccc}\mathrm{<span\; class="notranslate">\; 0.05</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0.05</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0.05</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0.05</span>}\end{array}\right]$

Step 3-3: Set the initial value of observation noise;

所以设定观测噪声初始值为：  According to the empirical data, it can be seen that the observation noise is more influential than the system noise, and v ₁ (k), v ₂ (k) satisfy independent and identical distribution, set

Therefore, the initial value of the observation noise is set as:

$\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; [</span>\mathrm{<span\; class="notranslate">\; V</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; V</span>}}^{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span><span\; class="notranslate">\; =</span>\left[\begin{array}{cc}\mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}\end{array}\right]$

Step 3-4: Set the observed noise autocorrelation value;

In order to track rapid convergence, the observation noise autocorrelation value is set to 100, and the initial value is set as follows:

$\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 100</span>}<span\; class="notranslate">\; \&times;</span>\left[\begin{array}{cccc}\mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}\end{array}\right]$

Step 4: Use 20 groups of Kalman filters, all using the initial values set in step 3, complete an iterative operation of the Kalman filter according to the Kalman filter iterative update equation defined in steps 2-4, and output a set of lanes The optimal estimated value of the line (ρ _i , θ _i )i=1, 2, 3, ..., 20

Step 5: Perform enable control on a single filter. the

Step 5-1: Define a lane line detected in the road image as (ρ, θ), and obtain (ρ, θ) to (ρ _i , θ _i ) i=1, 2, 3, ..., a distance D _i of 20

D _i ＝|ρ-ρ _i |+|θ-θ _i |*W _image

Where W _image represents the pixel width of the road image.

Step 5-2: Calculate the minimum distance D _min =min(D _i ), and record the minimum distance value of the last iteration as D _min ′.

Step 5-3: The enable counter corresponding to each group of filters is denoted as C _i , and satisfies the following formula

That is, when the minimum distance is less than or equal to the minimum distance of the previous iteration, the count value is increased by 1, otherwise the count value is decreased by 1. the

Step 5-4: Track the lane line, set the effective sample input number to 25, that is, when the same lane line is detected in 25 consecutive road images, the optimal estimate of the Kalman filter is considered valid, otherwise it is considered correct The number of samples of is insufficient, so no tracking is performed, so C _i passes through the threshold [0, 25], that is, C _i is processed according to the following formula:

${\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\mathrm{<span\; class="notranslate">\; 25</span>}& {\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; \&Greater\; Equal;</span>\mathrm{<span\; class="notranslate">\; 25</span>}\\ {\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}& {\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; 25</span>}\end{array}\right.$

Step 5-5: Get the enable E _i of each filter as follows:

${\mathrm{<span\; class="notranslate">\; E.</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\mathrm{<span\; class="notranslate">\; 1</span>}& {\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 25</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& {\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; 25</span>}\end{array}\right.<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1,2,3</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 20</span>}$

Step 6: Take the optimal estimated value output by the Kalman filter as the detection result. the

Using the optimal estimated value of the enable bit E _i =1 of the filter bank as the detection result instead of the detection value can make the detection result of the lane line in the lane departure warning system more stable, thereby greatly reducing the impact caused by vehicle body shaking, illumination changes, Inaccurate or missed detection of lane lines caused by road damage and other reasons.

The innovation of the present invention is:

1. The concept of filter group is adopted for the tracking of lane lines. According to the number of filters set, the corresponding number of lane lines can be tracked at the same time. Compared with the single tracking of the area of interest for lane lines, it greatly improves the The accuracy of lane line detection and anti-interference ability. the

2. The enable bit control of a single filter in the filter bank is added, and the filter corresponding to the undetected lane line is therefore in a non-working state, which greatly improves the calculation speed of the tracking algorithm. the

Description of drawings

Fig. 1 is a schematic flow chart of the present invention. the

Fig. 2 is a diagram of the number of matches between the output value of the filter bank and the detection value (C _i after passing through the [0, 25] valve), and the road image at this time can be known by the value of C _i in (a), (b) and (c) Two lane lines are detected in .

Figure 3 is a curve diagram of ρ detection value, optimal estimated value, and predicted value within t=30s in the test video. the

Detailed ways

The present invention adopts above steps, utilizes the Video And Image Tool Box among the Matlab R2009 (a), has realized whole algorithm. A section of road image was selected for testing in a high-speed section, and the Kalman filter bank was used for tracking. The accuracy of lane line detection was increased by 60%, and the average tracking time of each road image was only 0.08s, which achieved very ideal detection results. the

It should be noted that the following parameters do not affect the generality of this patent. the

1. The number of filters in the filter bank indicates the maximum number of lane lines that can be tracked at the same time. This algorithm is set to 20, but it can be increased or decreased according to the actual situation. the

2. The Kalman filter enables judgment, and the effective count value of the sample indicates the basic requirement for the continuous validity of the sample to output the optimal estimated value. The value of this algorithm is 25, but it can be increased or decreased according to actual needs. the

Claims (1)

1. the Multi-lane Lines method for tracing based on group of Kalman filters, comprises the following steps:

Step 1: in Vehicle Driving Cycle process, adopt Vehicular video collecting device collection vehicle road ahead image; And by the road image data input computing machine collecting, carry out lane detection;

Use Hough change detection to go out lane line, by lateral distance ρ and azimuth angle theta, represent detected lane line, i.e. (ρ _i, θ _i), i=1,2,3 ..., represent detected lane line on road image;

Step 2: set up the kalman filter models that lane line is followed the trail of;

Step 2-1: definition lane line state;

Because the motion of vehicle in two continuous frames video image acquisition process can be considered uniform motion, for the state of the lane line detecting in each frame road image, use state vector X (k) to represent:

X(k)=[ρ(k)，θ(k)，u(k)，ω(k)] ^T

Wherein represent that vehicle is at state k, i.e. t radial velocity constantly,

represent that vehicle is at state k, i.e. t angular velocity constantly;

Step 2-2: the state equation of Kalman filter in definition lane line tracing process;

Because system meets following formula by state X (k) to NextState X (k+1):

The system noise of set condition k is: W (k)=[w ₁(k), w ₂(k), w ₃(k), w ₄(k)] ^t, so the state equation of Kalman filter is expressed as:

X(k)=AX(k-1)+W(k)

Wherein

Step 2-3: the observation equation of definition lane line tracing process;

Because observed reading is the result of actual condition value and noise effect, observed reading can be expressed as follows:

Therefore obtaining observation equation is:

Z(k)=HX(k)+V(k)

Wherein measure matrix v (k) is observation noise and V (k)=[v ₁(k), v ₂(k)] ^t;

Step 2-4: definition Kalman filter iteration renewal equation:

X(k|k)=X(k|k-1)+Kg(k)(Z(k)-HX(k|k-1))

Wherein:

Kg(k)=P(k|k-1)H ^T/(HP(k|k-1)H ^T+R)

P(k|k)=(I-Kg(k)H)P(k|k-1)

X (k|k) is the optimum estimation value of the k state of Kalman filter output, X (k|k-1) is for utilizing the result of the k state that k-1 status predication goes out, Kg (k) is Kalman filter gain under state k, and Z (k) is the observed reading of state k, and H measures matrix;

Step 3: system initialization;

Step 3-1: set condition vector initial value;

From Kalman filter ultimate principle, as long as original state meets E[X (0)]=E[X (0|0)], resulting estimated value is without inclined to one side, so initialization system original state is null vector,

X(0)=[0?0?0?0] ^T

Step 3-2: initialization system noise initial value;

System noise W (k) is zero-mean stochastic process vector, and not in the same time under state system noise be to add up independently, its correlation matrix meets following formula:

Suppose it between different conditions noise, is also to add up independently in the same time, so matrix Q (k) is diagonal matrix; Because system noise is because vehicle mobility can cause, in tracing process, not quite, for convenience of calculation, rule of thumb data are supposed w in impact ₁(k), w ₂(k), w ₃(k), w ₄(k) independent same distribution, and meet

therefore initialization system noise initial value is:

Step 3-3: set observation noise initial value;

Rule of thumb the known observation noise of data is larger than system noise impact, and v ₁(k), v ₂(k) meet independent same distribution, set so set observation noise initial value, be:

Step 3-4: set observation noise autocorrelation value;

In order to follow the trail of rapid convergence, observation noise autocorrelation value is taken as 100, sets initial value as follows:

Step 4: adopt 20 groups of Kalman filter, all adopt the initial value of setting in step 3, according to the Kalman filter iteration renewal equation of step 2-4 definition, complete the interative computation of a Kalman filter, export the optimal estimation value (ρ of one group of lane line _i, θ _i), i=1,2,3 ..., 20;

Step 5: single filter is enabled to control;

Step 5-1: the lane line detecting in road image is defined as to (ρ, θ), obtains (ρ, θ) to (ρ by following formula respectively _i, θ _i), i=1,2,3 ..., 20 distance D _i

D _i=|ρ-ρ _i|+|θ-θ _i|*W _image

W wherein _imagethe pixel wide that represents road image;

Step 5-2: obtain minor increment D _min=min (D _i), and remember that the distance minimum value of last iteration is D _min';

Step 5-3: every group of enable counter corresponding to wave filter is designated as C _i, and meet following formula

When minor increment is less than or equal to the minor increment of last iteration, count value adds 1, otherwise count value subtracts 1;

Step 5-4: lane line is followed the trail of, setting effective sample input number is 25,, when continuous 25 width road images detect same lane line, thinks that the optimal estimation of Kalman filter is effective, otherwise the sample number praising is not enough, and does not follow the trail of; Therefore by C _iby thresholding [0,25], by following formula, process C _i:

Step 5-5: the E that enables that obtains each wave filter _ibe expressed as follows:

Step 6: using the optimal estimation value of Kalman filter output as testing result.

Images