Self-adaptive vehicle lighting system based on machine vision and adjusting method
By using machine vision and deep learning algorithms in vehicle lighting systems, combined with YOLOv8 detection algorithms, identifying elements in the road and surrounding environment, the problem that existing systems cannot fully consider vehicle speed and vehicle factors in front are solved, and more accurate Lighting adjustments improve safety and comfort at night.
Patent Information
- Application Number
- CN202410940779.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-15
- Publication Date
- 2024-11-26
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing vehicle lighting system cannot fully consider factors such as vehicle speed and the vehicle ahead during night driving, and there are blind spots and high costs.
Adaptive vehicle lighting system based on machine vision is adopted, combined with high-resolution cameras and deep learning algorithms, object detection and classification is carried out through the YOLOv8 algorithm, roads, vehicles, pedestrians and other elements are identified, and adapted lighting mode is automatically selected.
It improves the safety and comfort of night driving, reduces system costs, enhances the recognition ability of low-speed moving objects and non-reflective radar wave objects, and provides more accurate lighting adjustments.
Smart Images

Figure CN119018042A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to an adaptive vehicle lighting system and an adjustment method based on machine vision, and belongs to the field of intelligent control of vehicle lighting systems. Background Art
[0002] The adaptive vehicle lighting system is designed to intelligently adjust the brightness, direction and lighting pattern of the vehicle's headlights according to the driving environment and vehicle status, improving the safety and comfort of night driving.
[0003] The lighting system of existing vehicles monitors the external environment through devices such as photosensitive sensors, radars and infrared sensors. Photosensitive sensors are not smart enough and only adjust based on the ambient brightness, and cannot fully consider other factors such as vehicle speed and vehicles in front. Radar and infrared sensor technology solutions have blind spots. There may be blind spots for some slow-moving objects or objects that do not reflect radar waves. High cost: Radar and infrared sensor technologies are relatively expensive. Summary of the invention
[0004] In view of the above problems, the present invention provides an adaptive vehicle lighting system and adjustment method based on machine vision, which adopts a combination of hardware and algorithms; it can more accurately identify factors such as roads, pedestrians, and vehicles, and realize intelligent lighting adjustment.
[0005] The present invention adopts the following technical solution:
[0006] The present invention provides an adaptive vehicle lighting system based on machine vision, the system comprising a high-resolution camera for capturing real-time images of roads and surrounding environments; the high-resolution camera uses an infrared sensor or a low-light level sensor to assist in enhancing nighttime image capture;
[0007] Computer vision systems used to process images captured by high-resolution cameras.
[0008] The lighting adjustment method of the adaptive vehicle lighting system based on machine vision described in the present invention has the following adjustment strategy:
[0009] S1, the high-resolution camera captures images of the road and surrounding environment in real time;
[0010] S2, deep learning algorithms combined with computer vision systems for object detection, classification and tracking;
[0011] S3, using deep learning models to analyze the perceived scene and identify key elements such as roads, vehicles, pedestrians, and road curvature;
[0012] S4. Automatically select an adapted vehicle lighting mode based on the factors.
[0013] 3. The lighting adjustment method of the adaptive vehicle lighting system based on machine vision according to claim 2 is characterized in that: the object detection, classification and tracking method steps using the YOLOv8 algorithm in step S2 are as follows:
[0014] 1) Based on the C2f module in the YOLOv8 algorithm, replace the first convolution of the C2f module with the SA-Net module;
[0015] 2) Feature grouping based on the obtained feature map,
[0016] First, for a given feature map X∈R C×W×H Group by channel dimension into g groups; the expression is:
[0017] X=[X1,X2,X3……X g ],X k =R (C / G)×W×H
[0018] Where C represents the number of channels, H and W represent the height and width respectively.
[0019] Then X k It is divided into two parts along the channel dimension, and the number of channels in each part is half of the original number of channels, that is, C / 2G;
[0020] One branch generates a channel attention map based on the relationship between channels, and the other branch generates a spatial attention map through spatial relations;
[0021] 3) Based on the generated channel attention map, the channel attention weighting is realized to obtain the channel attention weight X k1 ' is a category representation, and its calculation formula is as follows:
[0022] X' k1 =σ(F c (s))·X k1 =σ(W 1s +b1)·X k1 (3)
[0023] In the formula, σ represents the Sigmoid activation function, F c (x)=Wx+b(W,b∈R (C / 2G)×1×1 ) is a linear function, and s means the average pooling feature;
[0024] 4) Based on the generated spatial attention map, the spatial attention weight X' is obtained k2 , to achieve the weighted spatial attention, the calculation formula is as follows:
[0025] X' k2=σ(W2·GN(X k2 )+b2)·X k2 (4)
[0026] Where W2,b2 is obtained through training, W2,b2∈R (C / 2G)×1×1 ; GN represents the group normalization function;
[0027] 5) The feature X′ after the channel attention and spatial attention weighted in the above step k1 , X′ k2 Splicing is performed along the channel dimension to aggregate all sub-features; then the spliced features are subjected to channel shuffling operations so that features of different groups are staggered in the channel dimension to ensure information interaction among the sub-features.
[0028] In the lighting adjustment method of the adaptive vehicle lighting system based on machine vision described in the present invention, in step S3, the deep learning model YOLOv8 extracts features of the surrounding environment information collected by the high-resolution camera, and uses the improved YOLOv8 model to extract multi-scale feature maps using the SA-Net module, exchange and integrate information, detect objects in the image, generate candidate frames and classify them.
[0029] Beneficial Effects
[0030] The present invention provides an adaptive vehicle lighting system and adjustment method based on machine vision. The adaptive lighting system based on machine vision comprehensively uses sensor data fusion, deep learning algorithm and user interaction to improve the safety and comfort of driving under multiple working conditions. In practical applications, the reliability, real-time performance and user experience of the system are taken into consideration. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] Figure 1 The SA-Net structure diagram of the attention mechanism of the present invention;
[0032] Figure 2 This is the C2f_S structure diagram of the present invention. DETAILED DESCRIPTION
[0033] In order to make the purpose and technical solution of the embodiment of the present invention clearer, the technical solution of the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings of the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all of the embodiments. Based on the described embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0034] The adaptive lighting system of the present invention is installed on the self-driving car, and the specific implementation steps are as follows:
[0035] A machine vision-based adaptive vehicle lighting system, the system comprising a high-resolution camera for capturing real-time images of a road and surrounding environment; the high-resolution camera employs an infrared sensor or a low-light level sensor for assisting in enhancing nighttime image capture;
[0036] Computer vision systems used to process images captured by high-resolution cameras.
[0037] The lighting adjustment method of the adaptive vehicle lighting system based on machine vision, the adjustment strategy is as follows:
[0038] S1, the high-resolution camera captures images of the road and surrounding environment in real time;
[0039] S2, deep learning algorithms combined with computer vision systems (algorithms) for object detection, classification and tracking;
[0040] The YOLOv8 algorithm is used to improve the C2f module in the algorithm, and the first convolution of the C2f module is replaced by the SA-Net module, named C2f_S. This enables it to process feature maps of different scales at the same time and exchange and integrate information between feature maps, so that the model can better utilize multi-scale information.
[0041] from Figure 1 It can be seen that the entire attention mechanism is mainly composed of 4 parts:
[0042] (1) Feature Grouping
[0043] First, for a given feature map X∈R C×W×H , C represents the number of channels, H and W represent the height and width respectively. Then group them into g groups X = [X1, X2, X3...X g ],X k =R (C / G)×W×H Finally, X k It is divided into two parts along the channel dimension, and the number of channels in each part is half of the original number of channels, that is, C / 2G. One branch generates a channel attention map based on the relationship between channels, and the other branch generates a spatial attention map based on the spatial relationship.
[0044] (2) Channel Attention
[0045] First, adaptive average pooling is performed on this branch to reduce the feature map on each channel to a size of 1x1 to obtain channel statistics. Then, linear transformation is performed on the average pooled features to enhance feature representation. Finally, the linearly transformed feature map is multiplied by the weight processed by the Sigmoid activation function to achieve channel attention weighting, thereby obtaining the channel attention weight X k1 ' is a category representation, and its calculation formula is as follows:
[0046] X' k1 =σ(F c (s))·X k1 =σ(W 1s +b1)·X k1 (5)
[0047] In the formula, σ represents the Sigmoid activation function, F c (x)=Wx+b(W,b∈R (C / 2G)×1×1 ) is a linear function. s means average pooling feature.
[0048] (3) Spatial Attention
[0049] First, group normalization is performed on this branch to obtain spatial statistical information, and then the normalized features are linearly transformed. Then, the feature map after linear transformation is multiplied by the weight processed by the Sigmoid activation function to embed the global information to obtain the spatial attention weight X' k2 , to achieve the weighted spatial attention. The calculation formula is as follows:
[0050] X' k2 =σ(W2·GN(X k2 )+b2)·X k2 (6)
[0051] Where W2,b2 is obtained through training, W2,b2∈R (C / 2G)×1×1 GN represents the group normalization function.
[0052] (4) Aggregation
[0053] First, the feature X′ after channel attention and spatial attention weighting k1 , X′ k2 Splicing is performed along the channel dimension to aggregate all sub-features. Then the spliced features are subjected to channel shuffling operation so that the features of different groups are staggered in the channel dimension to ensure the information interaction of each sub-feature.
[0054] Replace the first convolution of the C2f module with the SA-Net module, as Figure 2 As shown in the figure, the modified C2f module is renamed as C2f_S structure. This method allows the model to focus on the spatial information of the image and enhance the important areas in the feature map, thereby improving the model's ability to recognize targets. This makes the model more accurate and efficient in extracting vehicle information in foggy environments.
[0055] S3, using deep learning models to analyze the perceived scene and identify key elements such as roads, vehicles, pedestrians, and road curvature;
[0056] The surrounding environment information is collected through a high-resolution camera and passed into the improved deep learning model YOLOv8. First, the input data is preprocessed, including image normalization, resizing, noise removal and other steps. Next, feature extraction is performed, and the improved YOLOv8 model uses the SA-Net module to extract multi-scale feature maps for information exchange and integration. Then, the improved YOLOv8 model detects objects such as roads, vehicles, pedestrians, etc. in the image, generates candidate boxes and classifies them. The detected objects are further analyzed to identify key elements such as road curvature, lane lines, and traffic signs. The detection results are integrated to analyze the current road conditions and environmental information to generate visual scene understanding results.
[0057] S4. Automatically select the vehicle lighting mode that is adapted through the elements. The vehicle lighting adjustment has a visualization effect with interactive functions.
[0058] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.
Claims
1. An adaptive vehicle lighting system based on machine vision, characterized in that: The system includes a high-resolution camera for capturing real-time images of the road and surrounding environment; the high-resolution camera uses an infrared sensor or a low-light level sensor to assist in enhancing night-time image capture; Computer vision systems used to process images captured by high-resolution cameras.
2. A lighting adjustment method for an adaptive vehicle lighting system based on machine vision, characterized in that: The adjustment strategy is as follows: S1, the high-resolution camera captures images of the road and surrounding environment in real time; S2, deep learning algorithms combined with computer vision systems for object detection, classification and tracking; S3, using deep learning models to analyze the perceived scene and identify key elements such as roads, vehicles, pedestrians, and road curvature; S4. Automatically select an adapted vehicle lighting mode based on the factors.
3. The lighting adjustment method of the machine vision-based adaptive vehicle lighting system according to claim 2, characterized in that: The object detection, classification, and tracking method steps using the YOLOv8 algorithm in step S2 are as follows: 1) Based on the C2f module in the YOLOv8 algorithm, replace the first convolution of the C2f module with the SA-Net module; 2) Feature grouping based on the obtained feature map, First, for a given feature map X∈R C×W×H Group by channel dimension into g groups; the expression is: X=[X1,X2,X3……X g ],X k =R (C / G)×W×H Where C represents the number of channels, H and W represent the height and width respectively. Then X k It is divided into two parts along the channel dimension, and the number of channels in each part is half of the original number of channels, that is, C / 2G; One branch generates a channel attention map based on the relationship between channels, and the other branch generates a spatial attention map through spatial relations; 3) Based on the generated channel attention map, the channel attention weighting is realized to obtain the channel attention weight X k1 ' is a category representation, and its calculation formula is as follows: X' k1 =σ(F c (s))·X k1 =σ(W 1s +b1) X k1 (1) In the formula, σ represents the Sigmoid activation function, F c (x)=Wx+b(W,b∈R (C / 2G)×1×1 ) is a linear function, and s means the average pooling feature; 4) Based on the generated spatial attention map, the spatial attention weight X' is obtained k2 , to achieve the weighted spatial attention, the calculation formula is as follows: X' k2 =σ(W2·GN(X k2 )+b2)·X k2 (2) Where W2,b2 is obtained through training, W2,b2∈R (C / 2G)×1×1 ; GN represents the group normalization function; 5) The feature X′ after the channel attention and spatial attention weighted in the above step k1 , X′ k2 Splicing is performed along the channel dimension to aggregate all sub-features; then the spliced features are subjected to channel shuffling operations so that features of different groups are staggered in the channel dimension to ensure information interaction among the sub-features.
4. The lighting adjustment method of the adaptive vehicle lighting system based on machine vision according to claim 2 or 3, characterized in that: In step S3, the deep learning model YOLOv8 extracts features from the surrounding environment information collected by the high-resolution camera, and uses the improved YOLOv8 model to extract multi-scale feature maps using the SA-Net module, exchange and integrate information, detect objects in the image, generate candidate boxes, and classify them.
Citation Information
Cited By
Highway vehicle detection method and system based on computer vision
CN119625664A
