Infrared and visible light image fusion enhancement method and system for vehicle detection

Abstract

The present application relates to the technical field of image processing, in particular to a vehicle detection method and system for infrared and visible light image fusion enhancement, wherein the method comprises: obtaining respective base layer coefficients and detail layer coefficients by performing sub-window variance filtering decomposition on input infrared images and visible light images; performing visual saliency map processing on the infrared images and the visible light images to obtain corresponding saliency maps; performing weighted average fusion processing guided by the saliency maps to obtain base layer fusion coefficients; processing the detail layer coefficients to obtain a modified Laplacian operator and inputting the modified Laplacian operator into a dynamic threshold neural P system to obtain high-frequency coefficients after detail layer fusion; and performing sub-window variance filtering reconstruction according to the base layer fusion coefficients and the high-frequency coefficients to obtain a fusion image result after glare elimination, thereby solving the problem of glare interference on drivers. The present application can improve the quality and effect of the fusion image result, is simple to set up, and has stronger universality.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to an infrared and visible light image fusion enhancement method and system for vehicle detection. BACKGROUND

[0002] Traffic accidents are mostly related to cars, and there are many reasons for cars to cause traffic safety accidents, among which the incorrect use of light is particularly prominent, especially the light use habit of the driver has a great influence on the safety of driving on the night road, and the incorrect use of high beam will cause the driver of the oncoming vehicle to be disturbed by glare, causing the driver's line of sight to enter the visual blind area, and unable to see the pedestrians between the two vehicles and the things on the two sides and behind the front vehicle, which is easy to induce traffic accidents. According to the statistics of the Ministry of Public Security, among the traffic accidents occurring at night in China, the improper use of high beam is related to 30% to 40%. Therefore, the research on the anti-glare problem of night car has strong practicality and urgency.

[0003] In the research of anti-glare problem, an image acquisition device is usually used to collect images to provide the image information of the current road in front of the driver disturbed by glare. The commonly used image acquisition devices include infrared sensors and visible light sensors; among them, the infrared sensor does not need to rely on external environmental light, and itself emits infrared light for night vision imaging, with wide night vision range and less environmental influence, but it has certain defects in reflecting the real scene, and the resolution of the formed infrared image is low, and the signal-to-noise ratio is low; the visible light sensor can clearly reflect the detail information of the scene under certain conditions, but its imaging is easily affected by illumination, weather and other natural conditions. Therefore, researchers use image fusion methods to excavate the characteristic information of each source image according to the complementary nature of the advantages and disadvantages of the two, and then highlight the thermal target information to improve the understanding of the scene information by the visual system, so as to achieve the purpose of identifying camouflage and night vision.

[0004] At present, the fusion methods of infrared image and visible light image mainly include three categories: image fusion method based on multi-scale analysis, image fusion method based on sparse representation and image fusion method based on deep learning. Among them, the image fusion method based on multi-scale analysis is to transform the source image from spatial domain to transform domain for representation, then set the corresponding fusion rule in the transform domain for fusion, and finally reconstruct to obtain the fusion image result; the image fusion method based on sparse representation is to first learn an overcomplete dictionary from a large number of high-quality natural images, then try to represent the test sample with the least elements in the overcomplete dictionary, and finally reconstruct to obtain the fusion image result; the key lies in the construction of overcomplete dictionary, sparse coding and fusion rule; the image fusion method based on deep learning is to automatically extract the features of the input source image by constructing a deep neural network model, and under the constraint of the loss function, the ideal fusion image result is output after multiple training and optimization.

[0005] While all the aforementioned fusion methods can fuse infrared and visible light images, they all have certain drawbacks. Specifically, image fusion methods based on multi-scale analysis typically have complex fusion rules, and their transformation basis functions are usually only applicable to a specific type of image, resulting in poor universality. Fusion methods based on sparse representation usually divide the source image into image patches, but this step ignores the correlation between different image patches, leading to the loss of detailed information and affecting the quality of the fused image. Although image fusion methods based on deep learning have strong universality, their parameter settings are complex, and their overall fusion quality and effect are not as good as the previous two methods.

[0006] In addition to its strong practical significance for nighttime road traffic safety, anti-halo technology based on image enhancement and image fusion also has important application value in other key issues. For example, with the acceleration of my country's industrialization, the problem of smog has become increasingly prominent due to the large consumption of energy, affecting daytime road traffic and security monitoring. Image enhancement and fusion are important research directions for image dehazing, so anti-halo technology based on image enhancement and image fusion has a very broad application prospect.

[0007] Therefore, there is an urgent need for an infrared and visible light image fusion enhancement method for vehicle detection to solve the problem of anti-halo technology based on image enhancement and image fusion. This method can improve the quality and effect of the fused image results, and is simple to set up with greater versatility. Summary of the Invention

[0008] One of the objectives of this invention is to provide an infrared and visible light image fusion enhancement method for vehicle detection, which can improve the quality and effect of the fused image results. It is simple to set up and has greater versatility, in order to solve the problem that drivers are affected by glare, causing their vision to enter the blind spot, making it impossible to see pedestrians in the two vehicles and things on both sides and behind the vehicle in front, which can easily lead to traffic accidents.

[0009] The basic solution provided by this invention is: an infrared and visible light image fusion enhancement method for vehicle detection, comprising the following:

[0010] S1. Perform sub-window variance filtering decomposition on the input infrared and visible light images to obtain the base layer coefficients and detail layer coefficients of the infrared and visible light images, respectively.

[0011] S2. Perform visual saliency mapping on the infrared and visible light images to obtain the saliency maps of the infrared and visible light images respectively.

[0012] S3. Based on the base layer coefficients and saliency map, perform weighted average fusion processing to obtain the base layer fusion coefficients of the fused base layer;

[0013] S4. Process the detail layer coefficients of the infrared image and the visible light image respectively, and obtain the modified Laplacian operator of their detail layer coefficients;

[0014] S5. Input the modified Laplacian operator values ​​of the infrared image and the visible light image into the dynamic threshold neural P system respectively to obtain the high-frequency coefficients after the fusion of detail layer coefficients.

[0015] S6. Perform sub-window variance filtering and reconstruction based on the base layer fusion coefficients and high-frequency coefficients to obtain the final fused image result after halo removal.

[0016] Further, S1 includes:

[0017] S101, Input infrared image and visible light images ;

[0018] S102, respectively for and Perform sub-window variance filtering decomposition to obtain and Base layer coefficients and detail layer coefficients;

[0019] The sub-window variance filtering decomposition uses the sub-window variance filtering function: ;

[0020] in It is a sub-window variance filtering function; It is the source image. If... Perform sub-window variance filtering decomposition, then If for Perform sub-window variance filtering decomposition, then ; It is a filtered image; Indicates the window size of the filter; This represents the threshold variance value for preserving sharp edges;

[0021] For each one The base layer coefficient is:

[0022] ;

[0023] in It corresponds to the first The base layer coefficients of the layer source image, It is the first Layer filter window size, It is the first Layer threshold variance; and The smoothest image from the source image, as The base layer coefficient;

[0024] For each one The detailed layer coefficients are obtained through inequality relationships:

[0025] ;

[0026] in It is the first layer Detail layer coefficients.

[0027] Furthermore, the aforementioned ;

[0028] in , It is composed of pixels The local region defined by the filter centered on it. It is the first The intensity value of each pixel;

[0029] ,in , , It is a positive user parameter;

[0030] and , Indicates the number of decomposition layers.

[0031] Furthermore, the visual saliency map processing includes:

[0032] ;

[0033] in, Representing an image The salience map, Representing an image The total number of pixels in;

[0034] If two pixels have the same pixel intensity value, then their significance values ​​are equal. ;

[0035] in, Indicates pixel intensity value; Indicates that the pixel intensity value is equal to The number of pixels, For gray levels, the saliency of an infrared image saliency map of visible light image .

[0036] Furthermore, the weighted average fusion process includes: ;

[0037] Among them, weight ;

[0038] in, China The salience map, , China The salience map, .

[0039] Furthermore, the modified Laplace operator for the detail level coefficients for:

[0040] ;

[0041] in, Here is the weight matrix: ;

[0042] for:

[0043]

[0044] in, For located Detail layer coefficients for location.

[0045] Furthermore, the neurons in the dynamic threshold neural P system It is a neuron based on the pulsation mechanism. The state equation is:

[0046]

[0047]

[0048] in, For neurons From adjacent neurons Received peak value, For the corresponding local weights, As an external stimulus, for The peak value generated during ignition;

[0049] The peak value of the infrared image is: ;in A matrix of infrared images, Neurons in the P system The peak values ​​transmitted are accumulated.

[0050] The peak value of a visible light image is: ;in The matrix of the visible light image, Neurons in the P system The peak values ​​transmitted are accumulated.

[0051] based on and Obtain the high-frequency coefficients after fusing the detailed layer coefficients. :

[0052]

[0053] in, and For the first Layer position The detail layer coefficients at that location, For the first Layer position The high-frequency coefficients after fusing the detailed layer coefficients.

[0054] Furthermore, the fused image result for:

[0055] ;

[0056] in .

[0057] The second objective of this invention is to provide an infrared and visible light image fusion enhancement system for vehicle detection, which can improve the quality and effect of the fused image results, and is simple to set up and has greater versatility.

[0058] The present invention provides a second basic solution: an infrared and visible light image fusion enhancement system for vehicle detection, which adopts the above-mentioned infrared and visible light image fusion enhancement method for vehicle detection.

[0059] The basic principles and beneficial effects of this solution are as follows:

[0060] First, the input infrared and visible light images are decomposed by sub-window variance filtering to obtain the coefficients of the base layer and detail layer respectively. These coefficients have more concentrated energy and can better preserve the detailed information of the source image (infrared and visible light images) decomposition coefficients (base layer and detail layer coefficients), which is convenient for analysis and processing and helps to improve the accuracy of subsequent analysis and processing.

[0061] Then, visual saliency mapping is performed on the infrared and visible light images to obtain their corresponding saliency maps. Based on the saliency maps, a weighted average fusion process is performed on the base layer to obtain the base layer fusion coefficients, thereby effectively extracting the target contours from the source images and removing halo contours.

[0062] Next, the coefficients of the detail layer are processed to obtain their modified Laplacian operator. The obtained modified Laplacian operator values ​​are then input into the dynamic threshold neural P system to obtain the high-frequency coefficients after the detail layer fusion. This allows for the acquisition of better detail information in the source image, making the final fused image after dehalo removal easier for drivers to observe and recognize.

[0063] Finally, based on the fused base layer coefficients (base layer fusion coefficients) and detail layer coefficients (high frequency coefficients), sub-window variance filtering reconstruction is performed to obtain the final fused image result after de-halo, which is then viewed by drivers to solve the problem of drivers being interfered with by halos, causing their vision to enter blind spots, making it impossible to see pedestrians in the two lanes and things on both sides and behind the vehicle in front, which can easily lead to traffic accidents.

[0064] This solution not only better preserves the target contour information when removing halos, but also results in images with higher clarity and contrast, thus improving the quality and effect of the fused images. Furthermore, this solution does not require complex fusion rules and parameter settings, making the overall setup simple and more universal. Attached Figure Description

[0065] Figure 1 This is a schematic flowchart of an embodiment of an infrared and visible light image fusion enhancement method for vehicle detection according to the present invention;

[0066] Figure 2 This is a schematic diagram of an infrared image in an embodiment of an infrared and visible light image fusion and enhancement method for vehicle detection according to the present invention;

[0067] Figure 3 This is a schematic diagram of a visible light image in an embodiment of an infrared and visible light image fusion enhancement method for vehicle detection according to the present invention;

[0068] Figure 4 This is a schematic diagram of the base layer of the infrared image in an embodiment of an infrared and visible light image fusion enhancement method for vehicle detection according to the present invention;

[0069] Figure 5 This is a schematic diagram of the base layer of the visible light image in an embodiment of an infrared and visible light image fusion enhancement method for vehicle detection according to the present invention;

[0070] Figure 6 This is a schematic diagram of the detail layer of an infrared image in an embodiment of an infrared and visible light image fusion enhancement method for vehicle detection according to the present invention;

[0071] Figure 7 This is a schematic diagram of the detail layer of a visible light image in an embodiment of an infrared and visible light image fusion enhancement method for vehicle detection according to the present invention;

[0072] Figure 8 This is a schematic diagram of an infrared image used in a comparative experiment of an embodiment of an infrared and visible light image fusion enhancement method for vehicle detection according to the present invention.

[0073] Figure 9 This is a schematic diagram of a visible light image used in a comparative experiment of an embodiment of an infrared and visible light image fusion enhancement method for vehicle detection according to the present invention.

[0074] Figure 10 This is a schematic diagram of the fusion image results of the present invention in a comparative experiment of an infrared and visible light image fusion enhancement method for vehicle detection.

[0075] Figure 11 This is a schematic diagram of the image fusion result of the CNN method in a comparative experiment of an embodiment of the infrared and visible light image fusion enhancement method for vehicle detection of the present invention;

[0076] Figure 12 This is a schematic diagram of the image fusion result of the CONV_SR method in a comparative experiment of an embodiment of the infrared and visible light image fusion enhancement method for vehicle detection according to the present invention;

[0077] Figure 13 This is a schematic diagram of the LatLRR method fusion image results in a comparative experiment of an infrared and visible light image fusion enhancement method for vehicle detection according to the present invention. Detailed Implementation

[0078] The following detailed description illustrates the specific implementation method:

[0079] Example 1

[0080] The basic implementation examples are as follows: Figure 1 As shown: An infrared and visible light image fusion enhancement method for vehicle detection, including the following:

[0081] S1. Perform sub-window variance filtering decomposition on the input infrared and visible light images to obtain the base layer coefficients and detail layer coefficients of the infrared and visible light images respectively.

[0082] Specifically, including:

[0083] S101, Input infrared image and visible light images ,like Figure 2 and Figure 3 As shown, Figure 2 for , Figure 3 for ;

[0084] S102, respectively for and Perform sub-window variance filtering decomposition to obtain and The base layer coefficients and detail layer coefficients, such as Figure 4 , Figure 5 , Figure 6 and Figure 7 As shown, where Figure 4 for , Figure 5 for , Figure 6 for , Figure 7 for ;

[0085] Sub-window variance filtering decomposition uses the sub-window variance filtering function: ;

[0086] in It is a sub-window variance filtering function; It is the source image. If... Perform sub-window variance filtering decomposition, then If for Perform sub-window variance filtering decomposition, then ; It is a filtered image; Indicates the window size of the filter; This represents the threshold variance value for preserving sharp edges;

[0087] The above The edge preservation process for an image is a linear blend of the input image and the filtered image; specifically, for images with pixel dimensions... Given an input image centered (i.e., source image), filtered image The calculation is as follows:

[0088]

[0089] in, , This represents the preservation factor for each image patch (input image), used to control the degree of contribution from the input image; This represents a general smoothing filter used to evaluate the average intensity value of the input image. In this embodiment, a box filter is selected. The above formula combines the intensity value of the filtered image with the intensity value of the input image and the average intensity value of the input image. Oversharpening can be eliminated through the above formula.

[0090] make ,but , specifically, , It is composed of pixels The local region defined by the filter centered on it. It is the first s The intensity value of each pixel.

[0091] The filtering window is divided into four equal-sized sub-windows, A, B, C, and D, based on a local statistical marginal model, to evaluate the variance of each sub-window. This uniform sampling strategy is adopted to achieve high-performance filtering by summing the area tables.

[0092] make It is the set of intensity variances evaluated over four sub-windows. Define a preservation factor for each image patch to represent the intensity variance of the entire window. The value is:

[0093]

[0094] in , , It is a positive user parameter;

[0095] Based on the second condition of the edge model (the requirement for contrast intensity), use This serves as a potential edge indicator for the image. Additionally, the minimum sub-window variance value is added to the denominator of the above formula. This is to facilitate edge preservation under the third condition of the edge model, which has three conditions primarily used to identify edge similarity features on local image patches using simple statistical information: the first condition is that the edge consists of two sets of adjacent pixels; the second condition is that these two sets of adjacent pixels have contrast intensity (high global variance); and the third condition is that if the two sets of adjacent pixels have similar intensity (low local variance within a sub-region), the edge is more prominent. If a distinct edge exists... , equal If the value is , then the entire image patch will be It is completely preserved at that time, so This represents the threshold variance value for obtaining sharp edges within a given window. Furthermore, this filtering model guarantees that when filtering, the intensity values ​​of two adjacent points on the gradient will only become closer, avoiding over-sharpening.

[0096] In conclusion, ,in , m This indicates the number of decomposition layers, i.e., the number of filtering iterations.

[0097] For each one The base layer is:

[0098] ;

[0099] in It corresponds to the first m The base layer coefficients of the layer source image, It is the first m Layer filter window size, It is the first m Layer threshold variance; and The smoothest image from the source image, as The base layer coefficients; the number of layers in the source image was determined through ablation experiments. In this embodiment, the number of layers in the source image was determined to be 4 through ablation experiments, and the above... and Also, based on the ablation experiment settings, values ​​suitable for the current image are selected, and for each layer... and They may not be the same, in this embodiment , For each one The detailed layer coefficients are obtained through inequality relationships:

[0100] ;

[0101] in It is the first m layer The detail layer coefficients are obtained by subtracting the base layer coefficients from the source image.

[0102] S2. Perform visual saliency mapping on the infrared and visible light images to obtain the saliency maps of the infrared and visible light images respectively.

[0103] Specifically, visual saliency map processing includes:

[0104] ;

[0105] in, Representing an image The saliency map, i.e. The significance value at point, Representing an image The number of all pixels, Indicating the coefficients of the basic layer Intensity value at the point;

[0106] If two pixels have the same pixel intensity value, then their significance values ​​are equal. ;

[0107] in, Indicates pixel intensity value; Indicates that the pixel intensity value is equal to The number of pixels, In this embodiment, the grayscale level is specified. salient image of infrared image saliency map of visible light image ;

[0108] S3. Based on the base layer coefficients and saliency map, perform weighted average fusion processing to obtain the base layer fusion coefficients (i.e., the base layer fusion: the base layer coefficients for fusing infrared and visible light images).

[0109] Specifically, the base layer fusion coefficient for:

[0110] ;

[0111] Among them, weight ;

[0112] in, China The saliency map can be represented as , China The saliency map can be represented as .

[0113] S4. Process the detail layer coefficients of the infrared image and the visible light image respectively, and obtain the modified Laplacian operator of their detail layer coefficients;

[0114] Specifically, the modified Laplacian operator for calculating the coefficients of the detail level. :

[0115] ;

[0116] in, Here is the weight matrix: ;

[0117] This is a detail edge operator used to calculate the detail edge information of a certain region of an image.

[0118]

[0119] in, For located The detail layer coefficient of the location, Position representation in the image (detail layer coefficients) OK The pixels in the column.

[0120] S5. Input the modified Laplacian operator values ​​of the infrared image and the visible light image into the dynamic threshold neural P system respectively to obtain the high-frequency coefficients after fusing the detail layer coefficients (i.e., fusing the detail layer: fusing the detail layer coefficients of the infrared image and the visible light image).

[0121] Specifically, the dynamic threshold neural P system is a set neural network model, and the neurons in the dynamic threshold neural P system are... It is a neuron based on the pulsation mechanism. The state equation is:

[0122]

[0123]

[0124] in, For neurons From adjacent neurons Received peak value, For the corresponding local weights (i.e., neurons) its neighboring neurons The connection weight values ​​between them, where and It also represents the point in the corresponding row and column. As an external stimulus, for The peak value generated during ignition; the peak value in the infrared image is: ;in A matrix of infrared images, Neurons in the P system The peak values ​​transmitted are accumulated. express time The state of the neuron at the location; express time Position neuron state; Indicates the peak state of the neuron; express Neighborhood; Indicates the peak value of the neuron threshold; express time Neuron threshold at location; express time Threshold of location neurons.

[0125] The peak value of a visible light image is: ;in The matrix of the visible light image, Neurons in the P system The peak values ​​transmitted are accumulated.

[0126] based on and Obtain the high-frequency coefficients after fusing the detailed layer coefficients. :

[0127]

[0128] in, and For the first Layer position The detail layer coefficients at that location, For the first Layer position The fusion result of the detailed layer coefficients (the fused high-frequency coefficients).

[0129] S6. Perform sub-window variance filtering and reconstruction based on the base layer fusion coefficients and high-frequency coefficients to obtain and output the final fused image result after halo removal.

[0130] Specifically, the fused image results for:

[0131]

[0132] in .

[0133] This embodiment also provides an infrared and visible light image fusion enhancement system for vehicle detection, which employs the above-described infrared and visible light image fusion enhancement method for vehicle detection.

[0134] This scheme first decomposes the input infrared and visible light images by performing sub-window variance filtering to obtain the coefficients of their respective base layer and detail layer. These coefficients have more concentrated energy, which can better preserve the detailed information of the source image (infrared and visible light images) decomposition coefficients (base layer and detail layer coefficients), making it easier to analyze and process, and improving the accuracy of subsequent analysis and processing.

[0135] Then, visual saliency mapping is performed on the infrared and visible light images to obtain their corresponding saliency maps. Based on the saliency maps, a weighted average fusion process is performed on the base layer to obtain the base layer fusion coefficients, thereby effectively extracting the target contours from the source images and removing halo contours.

[0136] Next, the coefficients of the detail layer are processed to obtain their modified Laplacian operator. The obtained modified Laplacian operator values ​​are then input into the dynamic threshold neural P system to obtain the high-frequency coefficients after the detail layer fusion. This allows for the acquisition of better detail information in the source image, making the final fused image after dehalo removal easier for drivers to observe and recognize.

[0137] Finally, based on the fused base layer coefficients (base layer fusion coefficients) and detail layer coefficients (high frequency coefficients), sub-window variance filtering reconstruction is performed to obtain the final fused image result after de-halo, which is then viewed by drivers to solve the problem of drivers being interfered with by halos, causing their vision to enter blind spots, making it impossible to see pedestrians in the two lanes and things on both sides and behind the vehicle in front, which can easily lead to traffic accidents.

[0138] This solution not only better preserves the target contour information when removing halos, but also has higher clarity and contrast in the fused image results, making it more effective. Compared with existing technologies, it can improve the quality and effect of fused image results. Moreover, this solution does not require complex fusion rules and parameter settings, and the overall setup is simple, making it more universal.

[0139] To verify the above effects, comparative experiments were conducted, comparing this scheme with existing technologies. Specifically, the scheme was compared with the classic CNN fusion method (CNN), the Conv_SR method (Conv_SR), and the LatLRR method (LatLRR); where CNN is a deep learning-based image fusion method, Conv_SR is a sparse representation-based image fusion method, and LatLRR is a multi-scale analysis-based image fusion method. The parameter settings for the above comparison methods refer to existing technologies.

[0140] enter Figure 8 and Figure 9 The infrared and visible light images shown are fused and enhanced using the proposed scheme, CNN fusion method, Conv_SR method, and LatLRR method, respectively. The fused image results are output as follows: Figure 10 , Figure 11 , Figure 12 ,and Figure 13 As shown.

[0141] As can be seen from the fused image results, the method of this patent has a significantly better effect in removing halos, and the fused image results also have better clarity and contrast. Compared with existing technologies, this solution can improve the quality and effect of fused image results.

[0142] The above descriptions are merely embodiments of the present invention. Commonly known structures and characteristics of the solutions are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A method for enhancing infrared and visible light images for vehicle detection, characterized in that, Includes the following: S1. Perform sub-window variance filtering decomposition on the input infrared and visible light images to obtain the base layer coefficients and detail layer coefficients of the infrared and visible light images respectively. S2. Perform visual saliency mapping on the infrared and visible light images to obtain the saliency maps of the infrared and visible light images respectively. S3. Based on the base layer coefficients and saliency map, perform weighted average fusion processing to obtain the base layer fusion coefficients of the fused base layer; S4. Process the detail layer coefficients of the infrared image and the visible light image respectively, and obtain the modified Laplacian operator of their detail layer coefficients; S5. Input the modified Laplacian operator values ​​of the infrared image and the visible light image into the dynamic threshold neural P system respectively to obtain the high-frequency coefficients after the fusion of detail layer coefficients. S6. Perform sub-window variance filtering and reconstruction based on the base layer fusion coefficients and high-frequency coefficients to obtain the final fused image result after halo removal.

2. The infrared and visible light image fusion enhancement method for vehicle detection according to claim 1, characterized in that: S1 includes: S101, Input infrared image and visible light images ; S102, respectively for and Perform sub-window variance filtering decomposition to obtain and Base layer coefficients and detail layer coefficients; The sub-window variance filtering decomposition uses the sub-window variance filtering function: ; in It is a sub-window variance filtering function; It is the source image. If... Perform sub-window variance filtering decomposition, then If for Perform sub-window variance filtering decomposition, then ; It is a filtered image; Indicates the window size of the filter; This represents the threshold variance value for preserving sharp edges; For each one The base layer coefficient is: ； in It corresponds to the first The base layer coefficients of the layer source image, It is the first Layer filter window size, It is the first Layer threshold variance; and The smoothest image from the source image, as The base layer coefficient; For each one The detail layer coefficients are obtained through the following relationship: ； in It is the first layer Detail layer coefficients.

3. The infrared and visible light image fusion enhancement method for vehicle detection according to claim 2, characterized in that: The ; in , It is composed of pixels The local region defined by the filter centered on it. It is the first The intensity value of each pixel; ,in , , It is a positive user parameter; and , Indicates the number of decomposition layers.

4. The infrared and visible light image fusion enhancement method for vehicle detection according to claim 2, characterized in that: The visual saliency map processing includes: ； in, Representing an image The salience map, Representing an image The number of all pixels, I ( p ) represents the coefficient of the basic layer. p Intensity value at the point; If two pixels have the same pixel intensity value, then their significance values ​​are equal. ; in, Indicates pixel intensity value; Indicates that the pixel intensity value is equal to The number of pixels, For gray levels, the saliency of an infrared image saliency map of visible light image .

5. The infrared and visible light image fusion enhancement method for vehicle detection according to claim 4, characterized in that: The weighted average fusion process includes: ; Among them, weight ; in, China The salience map, , China The salience map, .

6. The infrared and visible light image fusion enhancement method for vehicle detection according to claim 5, characterized in that: The modified Laplace operator of the detail level coefficients for: ； in, Here is the weight matrix: ; for: in, For located Detail layer coefficients for location.

7. The infrared and visible light image fusion enhancement method for vehicle detection according to claim 6, characterized in that: Neurons in the dynamic threshold neural P system It is a neuron based on the pulsation mechanism. The state equation is: in, For neurons From adjacent neurons Received peak value, For the corresponding local weights, As an external stimulus, for The peak value generated during ignition; The peak value of the infrared image is: ;in A matrix of infrared images, Neurons in the P system The peak values ​​transmitted are accumulated. The peak value of a visible light image is: ;in The matrix of the visible light image, Neurons in the P system The peak values ​​transmitted are accumulated. based on and Obtain the high-frequency coefficients after fusing the detailed layer coefficients. : in, and For the first Layer position The detail layer coefficients at that location, For the first Layer position High-frequency coefficients after fusion of detailed layer coefficients; u ij ( t +1 indicates t The state of the neuron at position ij at time +1. u ij ( t )express t The state of the neuron at position ij at time 1. u Let δr be the peak state of the neuron, and δr be the set of all neurons in a neighborhood of radius r. τ The peak value of the neuron threshold. τ ij (t+1) represents t The neuron threshold at position ij at time +1.

8. The infrared and visible light image fusion enhancement method for vehicle detection according to claim 7, characterized in that: The fused image result for: ； in .

9. An infrared and visible light image fusion enhancement system for vehicle detection, characterized in that: The infrared and visible light image fusion enhancement method for vehicle detection as described in any one of claims 1-8 is adopted.

Images