Projection light blanket dynamic fitting algorithm of intelligent self-closed-loop vehicle lamp

Abstract

The invention discloses a projection light blanket dynamic fitting algorithm of an intelligent self-closed-loop vehicle lamp, and belongs to the field of intelligent traffic and intelligent light control. The method comprises the following steps: S1, capturing a road image containing a lane line through a camera; s2, processing the road image, and identifying a lane line; and S3, carrying out dynamic fitting on the identified lane line. According to the projection light blanket dynamic fitting algorithm of the intelligent self-closed-loop vehicle lamp, the fitting precision is high, the robustness is high, better driving experience can be brought to a driver, and the safety of night driving is improved.

Description

Technical Field

[0001] The present invention relates to a projection light carpet dynamic fitting algorithm for an intelligent self-closed-loop vehicle lamp, belonging to the fields of intelligent transportation and intelligent lighting control. Background Art

[0002] Currently, in the function of the intelligent high-definition vehicle lamp projection dynamic light carpet, the accurate recognition and fitting of lane lines are the key to ensuring the functional effect experience. Traditional methods often rely on simple mathematical models or fixed optimization strategies, such as the three-segment fitting method of lane lines, and it is difficult to adapt to complex and changeable road environments. Therefore, it is particularly important to develop a lane line fitting algorithm that can dynamically adapt to different road conditions and provide high fitting accuracy. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and provide a projection light carpet dynamic fitting algorithm for an intelligent self-closed-loop vehicle lamp, which has high fitting accuracy, strong robustness, can bring a better driving experience to drivers, and improve the safety of night driving.

[0004] To solve the above technical problem, the technical solution of the present invention is as follows:

[0005] A projection light carpet dynamic fitting algorithm for an intelligent self-closed-loop vehicle lamp, which includes the following steps:

[0006] Step S1: Capture a road image containing lane lines through a camera;

[0007] Step S2: Process the road image to identify the lane lines;

[0008] Step S3: Dynamically fit the identified lane lines.

[0009] Further, in the step S2, when processing the road image to identify the lane lines, it specifically includes the following steps:

[0010] Use a deep learning network to process the road image to identify the position and shape of the lane lines.

[0011] Further, in the step S3, when dynamically fitting the identified lane lines, it specifically includes the following steps:

[0012] Step S31: Establish a lane line equation and obtain the fitting error of the lane line equation;

[0013] Step S32: Use a probability model to guide the search process, and quickly locate the potential optimal solution area by continuously updating the probability estimate of the lane line fitting error;

[0014] Step S33: Continuously adjust the parameter vector of the lane line equation through iterative updates to minimize the fitting error. When the fitting error reaches a preset threshold or the number of iterations reaches the upper limit, stop the iteration and output the optimal fitting result.

[0015] Furthermore, in step S31, to establish the lane line equation and obtain the fitting error of the lane line equation, the following specific steps are included:

[0016] The expression of the lane line equation is:

[0017]

[0018] where g(x, w) is the fitting error of the lane line equation;

[0019] x is the position of the lane line in the road image;

[0020] w is the parameter vector, and the parameter vector contains the shape feature information and position feature information of the lane line;

[0021] φi(x) is the basis function;

[0022] n is the number of basis functions.

[0023] Furthermore, in step S32, use the probability model to guide the search process, and quickly locate the potential optimal solution area by continuously updating the probability estimate of the lane line fitting error. The specific steps are as follows:

[0024] First, define the lane line equation as a quadratic polynomial:

[0025] y = ax 2 + bx + c;

[0026] where a, b, and c are the coefficients of the quadratic polynomial;

[0027] The assumed function value f(w) of Gaussian process regression follows a Gaussian process, expressed as:

[0028] f(w) ~ GP(m(w), k(w, w'));

[0029] where m(w) is the mean function and k(w, w') is the covariance function;

[0030] Then, use the lane line equation as the mean function m(w) of Gaussian process regression, expressed as:

[0031] m(w) = ax 2 + bx + c;

[0032] where w = (x, y) is the observed data point;

[0033] Finally, an observation model is generated, and the expression of the observation model is:

[0034]

[0035] where f(w i ) is the value of the latent Gaussian process at w i , and ∈ i is independent and identically distributed noise;

[0036] When predicting the distribution, for the new parameter vector w * , the distribution of the predicted fitting error y * can be obtained. According to the formula of Gaussian process regression, the mean μ(w * ) and variance σ 2 (w * ) of the predicted distribution are obtained. The expressions for the mean μ(w * ) and variance σ 2 (w * ) of the predicted distribution are:

[0037] μ(w * ) = m(w * ) + k(w * ,W)(K(W,W) + σ 2 I) -1 (y - m(W))

[0038] σ 2 (w * ) = k(w * ,w * ) - k(w * ,W)(K(W,W) + σ 2 I) -1 k(W,w * );

[0039] where W is the set of observed parameter vectors;

[0040] y is the corresponding fitting error vector;

[0041] σ2 is the variance of the observation noise;

[0042] I is the identity matrix;

[0043] k(w * ,W) is the covariance vector between the new parameter vector w * and the observed parameter vector W;

[0044] K(W,W) is the covariance matrix between the observed parameter vectors W.

[0045] Further, in step S33, the parameter vector of the lane line equation is continuously adjusted through iterative update, which specifically includes the following steps:

[0046] In each iteration, Gaussian process regression is used to predict the fitting error under different parameter vectors, and the acquisition function is used to select the next parameter vector to be evaluated. Through continuous iterative update, the optimal parameter vector is gradually approximated.

[0047] Adopting the above technical solution, the present invention has the following beneficial effects:

[0048] 1. Through the Bayesian optimization algorithm, the present invention can dynamically adapt to different road conditions and achieve a relatively high-precision fitting of the lane line, so that the projection light carpet of the vehicle lamp has a better effect in a complex and changeable road environment.

[0049] 2. The dynamic fitting algorithm of the present invention has strong robustness to changes in the road environment and can cope with complex and changeable road scenes. Description of the Drawings

[0050] Figure 1 is a flowchart of the projection light carpet dynamic fitting algorithm of the intelligent self-closed-loop vehicle lamp of the present invention;

[0051] Figure 2 is a comparison diagram of the effects before and after optimization of the vehicle lamp projection of the present invention in a straight road condition;

[0052] Figure 3 is a comparison diagram of the effects before and after optimization of the vehicle lamp projection of the present invention in a curved road condition. Detailed Embodiments

[0053] In order to make the content of the present invention more clearly understood, the present invention will be further described in detail below according to specific embodiments and in conjunction with the accompanying drawings.

[0054] As Figure 1 shown, this embodiment provides a projection light carpet dynamic fitting algorithm for an intelligent self-closed-loop vehicle lamp, which includes the following steps:

[0055] Step S1: Capture a road image containing lane lines through a high-definition camera.

[0056] Step S2: Process the road image using a deep learning network to identify the position and shape of the lane lines.

[0057] Step S3: Dynamically fit the identified lane lines, which specifically includes the following steps:

[0058] Step S31: Establish a lane line equation and obtain the fitting error of the lane line equation, specifically:

[0059] The expression of the lane line equation is as follows:

[0060]

[0061] Among them, g(x, w) is the fitting error of the lane line equation, that is, the difference between the actual observed value and the model predicted value;

[0062] x is the position of the lane line in the road image (such as pixel coordinates);

[0063] w is the parameter vector, and the parameter vector contains the shape feature information and position feature information of the lane line;

[0064] φi(x) is the basis function, and the basis function is a quadratic polynomial, which is used to map the input position x of the lane line in the road image to the feature space;

[0065] n is the number of basis functions.

[0066] The parameter vector w of the above lane line equation is used as the input of the Gaussian process, and the fitting error will be used as the output. The parameter vector w is iteratively updated through the Bayesian optimization algorithm to minimize the fitting error.

[0067] Step S32: Use the probability model to guide the search process, and quickly locate the potential optimal solution area by continuously updating the probability estimate of the lane line fitting error. Specifically:

[0068] Assume that the fitting error g(x, w) is generated by a potential Gaussian process, which is defined by the mean function m(w) and the covariance function k(w, w′). Through these two functions, we can perform probability modeling on the fitting error and use the observed data to infer the properties of the Gaussian process, so as to realize the prediction and uncertainty evaluation of the lane line equation fitting error.

[0069] Combining Gaussian process regression (GPR) with the definition of the lane line equation, an observation model can be constructed to predict the fitting error of the lane line equation under different parameter combinations:

[0070] First, define the lane line equation as a quadratic polynomial:

[0071] y = ax 2 + bx + c;

[0072] Among them, a, b, and c are the coefficients of the quadratic polynomial.

[0073] The assumed function value f(w) of Gaussian process regression follows a Gaussian process, which is expressed as:

[0074] f(w) ∼ GP(m(w), k(w, w′));

[0075] Among them, m(w) is the mean function, and k(w, w′) is the covariance function (i.e., the kernel function); in particular, the mean function m(w) can be a constant, a linear function, or a more complex function, which represents the prior expectation of the fitting error when there is no observed data. In a simple case, we can assume that m(w) = 0, that is, it is considered that the expected value of the fitting error is zero in the absence of data. The covariance function k(w, w′) describes the correlation between the fitting errors under two different parameter vectors w and w′.

[0076] Then, the lane line equation is used as the mean function m(w) of the Gaussian process regression, expressed as:

[0077] m(w) = ax 2 + bx + c;

[0078] where w = (x, y) is the observed data point.

[0079] Finally, an observation model is generated, and the expression of the observation model is:

[0080]

[0081] where f(w i ) is the value of the latent Gaussian process at w i , and ∈ i is independent and identically distributed noise.

[0082] Suppose we have a set of observed data where w i is a sample of the parameter vector, and y i is the corresponding fitting error (possibly with noise).

[0083] When predicting the distribution, for a new parameter vector w * , the distribution of the predicted fitting error y * can be obtained. According to the formula of Gaussian process regression, the mean μ(w * ) and variance σ 2 (w * ) of the predicted distribution are obtained. The expressions for the mean μ(w * ) and variance σ 2 (w * ) of the predicted distribution are:

[0084] μ(w * ) = m(w * ) + k(w * , W)(K(W, W) + σ 2 I) -1 (y - m(W)); σ 2 (w *) = k(w * , w * ) - k(w * , W)(K(W, W) + σ 2 I) -1 k(W, w * );

[0085] Where W is a set of observed parameter vectors;

[0086] y is the corresponding μ(w * ) = m(w * ) + k(w * , W)(K(W, W) + σ 2 I) -1 (y - m(W))

[0087] σ2 is the observation noise σ 2 (w * ) = k(w * , w * ) - k(w * , W)(K(W, W) + σ 2 I) -1 k(W, w * )

[0088] I is the identity matrix;

[0089] k(w * , W) is the covariance vector between the new parameter vector w * and the observed parameter vector W;

[0090] K(W, W) is the covariance matrix between the observed parameter vectors W.

[0091] The mean μ(w * ) represents the expected value of the fitting error corresponding to the optimal parameter vector w * under the prediction of the Gaussian process regression model. The closer the mean μ(w * ) is to 0, the smaller the fitting error corresponding to the predicted optimal parameter vector w * , that is, the better the lane line fitting effect. On the contrary, the larger the mean μ(w * ), the larger the fitting error and the worse the fitting effect.

[0092] The variance σ 2 (w * ) represents the uncertainty of the fitting error corresponding to the optimal parameter vector w * under the prediction of the Gaussian process regression model. The smaller the variance σ 2 (w * ), the smaller the predicted optimal parameter vector w *The smaller the uncertainty of the corresponding fitting error, that is, the more reliable and stable the prediction result. On the contrary, σ 2 (w * ) The larger the variance, the greater the uncertainty, and the prediction result may be less reliable or less stable.

[0093] In the lane line fitting algorithm of this embodiment, the above Bayesian optimization algorithm is used to find the optimal parameter vector w * , so as to minimize the fitting error of the lane line equation.

[0094] Step S33: Continuously adjust the parameter vector of the lane line equation through iterative update to minimize the fitting error. When the fitting error reaches the preset threshold or the number of iterations reaches the upper limit, stop the iteration and output the optimal fitting result. Among them, continuously adjusting the parameter vector of the lane line equation through iterative update is specifically:

[0095] In each iteration, Gaussian process regression is used to predict the fitting error under different parameter vectors, and the acquisition function (such as expected improvement) is used to select the next parameter vector to be evaluated. Through continuous iterative update, the optimal parameter vector can be gradually approximated, so as to achieve accurate lane line fitting.

[0096] In practical applications, the above dynamic fitting algorithm can comprehensively balance the mean and variance according to the actual situation. Because a low mean indicates a small fitting error in the prediction, but if the variance is very large, it means that this prediction may be less stable or less reliable. On the contrary, even if the mean is slightly higher, but if the variance is very small, it means that the prediction result is relatively stable and reliable. Specifically, for the actual application scenario of the projection light carpet of the car headlight, it is required that the light cannot change frequently, so the stability requirement is relatively high, and it is more inclined to select a parameter vector with a smaller variance, even if its mean is slightly higher. Figure 2 State A in it represents the situation where the projection area of the light carpet is too narrow in the prior art, Figure 2 State B in it represents the situation where the projection area of the light carpet is too wide in the prior art, Figure 2 State C in it represents the situation after the projection area of the light carpet is optimized by the dynamic fitting algorithm. Figure 3 State C in it represents the situation where the projection area of the light carpet in the prior art cannot fully match the curve, Figure 3 State D in it represents the situation after the projection area of the light carpet in the curve is optimized by the dynamic fitting algorithm. By adopting the dynamic fitting algorithm of this embodiment, the projection light carpet of the car headlight can have a better effect in a complex and changeable road environment.

[0097] The specific embodiments described above further elaborate on the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A dynamic fitting algorithm for projection light blanket of intelligent self-closed-loop vehicle lamp, characterized in that: It includes the following steps: Step S1, capturing a road image including lane lines through a camera; Step S2: Process the road image to identify lane lines; Step S3: dynamically fit the identified lane lines.

2. The projection light blanket dynamic fitting algorithm of the intelligent self-closed loop vehicle lamp according to claim 1 is characterized in that: In step S2, the road image is processed to identify lane lines, which specifically includes the following steps: The deep learning network is used to process road images and identify the position and shape of lane lines.

3. The projection light blanket dynamic fitting algorithm of the intelligent self-closed loop vehicle lamp according to claim 1 is characterized in that: In step S3, the identified lane line is dynamically fitted, which specifically includes the following steps: Step S31, establishing a lane line equation, and obtaining a fitting error of the lane line equation; Step S32: using the probability model to guide the search process, and quickly locating the potential optimal solution area by continuously updating the probability estimate of the lane line fitting error; Step S33: Continuously adjust the parameter vector of the lane line equation through iterative updating to minimize the fitting error. When the fitting error reaches a preset threshold or the number of iterations reaches an upper limit, the iteration is stopped and the optimal fitting result is output.

4. The projection light blanket dynamic fitting algorithm of the intelligent self-closed loop vehicle lamp according to claim 3 is characterized in that: In step S31, a lane line equation is established to obtain a fitting error of the lane line equation, which specifically includes the following steps: The lane line equation is expressed as: Among them, g(x, w) is the fitting error of the lane line equation; x is the position of the lane line in the road image; w is a parameter vector, which contains the shape feature information and position feature information of the lane line; φi(x) is the basis function; n is the number of basis functions.

5. The projection light blanket dynamic fitting algorithm of the intelligent self-closed loop vehicle lamp according to claim 3 is characterized in that: In step S32, the probability model is used to guide the search process, and the potential optimal solution area is quickly located by continuously updating the probability estimation of the lane line fitting error, which specifically includes the following steps: First, define the lane line equation as a quadratic polynomial: y=ax 2 +bx+c; Where a, b, and c are the coefficients of the quadratic polynomial; The assumption of Gaussian process regression is that the function value f(w) obeys the Gaussian process, which is expressed as: f(w)~GP(m(w),k(w,w′)); Among them, m(w) is the mean function, k(w, w′) is the covariance function; Then, the lane line equation is used as the mean function m(w) of Gaussian process regression, expressed as: m(w)=ax 2 +bx+c; Where w = (x, y) is the observed data point; Finally, an observation model is generated, and the expression of the observation model is: Among them, f(w i ) is the underlying Gaussian process in w i The value at ∈ i is independent and identically distributed noise; When predicting the distribution, for the new parameter vector w * , we can get the prediction fitting error y * According to the Gaussian process regression formula, we can get the mean μ(w * ) and variance σ 2 (w * ), predict the mean μ(w * ) and variance σ 2 (w * ) is: μ(w * )=m(w * )+k(w * ,W(K(W,W)+σ 2 I) -1 (-m(W)) s 2 (oh * )=k(ω * Oh, oh * )-k(ω * ,W)(K(W,W)+σ 2 I) -1 k(W,ω * ); Where W is the set of observation parameter vectors; y is the corresponding fitting error vector; σ2 is the variance of the observation noise; I is the identity matrix; k(w * , W) is the new parameter vector w * The covariance vector between and the observed parameter vector W; K(W, W) is the covariance matrix between the observation parameter vector W.

6. The projection light blanket dynamic fitting algorithm of the intelligent self-closed loop vehicle lamp according to claim 3 is characterized in that: In step S33, the parameter vector of the lane line equation is continuously adjusted through iterative updating, which specifically includes the following steps: In each iteration, Gaussian process regression is used to predict the fitting error under different parameter vectors, and the next parameter vector to be evaluated is selected according to the acquisition function. Through continuous iterative updates, the optimal parameter vector is gradually approached.

Images