An adaptive high-beam anti-glare method and system

By using semantic segmentation and one-dimensional angle coding in the adaptive high beam lighting system, the problems of regional lighting loss and computational resource waste during target recognition in the prior art are solved, and more effective anti-glare control and resource optimization are achieved.

CN123104515APending Publication Date: 2026-11-17SAIC VOLKSWAGEN AUTOMOTIVE CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611621807.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-20
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing adaptive high-beam lighting systems have difficulty effectively distinguishing the anti-glare requirements of different areas when identifying targets, resulting in lighting loss in non-target areas or excessive consumption of computing resources.

Method used

The upstream perception controller performs semantic segmentation on the forward view image to generate a one-dimensional angle coding sequence. The downstream lighting controller controls the brightness of the matrix LED high beams based on the coding, and determines the occlusion depth by using semantic category priority and angle resolution, thereby reducing the downstream computing burden.

Benefits of technology

This approach effectively reduces the downstream computing burden while distinguishing the shading needs of different areas, avoiding lighting loss in non-target areas, and improving the anti-glare effect.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123104515A_ABST
    Figure CN123104515A_ABST
Patent Text Reader

Abstract

This invention provides an adaptive high-beam anti-glare method and system. An upstream perception controller performs semantic segmentation on the forward-view image, dividing it into horizontal angle segments according to angular resolution, obtaining the semantic category distribution of each angle segment, determining the occlusion depth based on the priority of the semantic categories, and projecting the two-dimensional pixel category labels along the horizontal angle direction into a one-dimensional angle-coded sequence where each code point corresponds to an angle segment, which is then transmitted via the vehicle bus. A downstream lighting controller determines the corresponding code point based on the channel pre-calibrated angle, the sequence starting angle, and the angular resolution, reads the occlusion depth, determines the target brightness according to the mapping relationship between depth and brightness, and controls the matrix LED high beams. This invention uses one-dimensional angle encoding to transmit semantic occlusion requirements, achieving a division of labor between perception analysis and channel control, reducing the downstream computational burden, and implementing differentiated occlusion according to angle segments.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of automotive intelligent lighting technology, specifically to an adaptive high-beam anti-glare method and system. Background Technology

[0002] When vehicles are driving at night, high beams can expand the visibility of the road ahead, but when they illuminate oncoming vehicles or other road users, they can cause glare. Adaptive high beams reduce glare by identifying targets ahead and adjusting the brightness of the corresponding illuminated areas, maintaining road illumination while minimizing glare. For high beams using matrix LED light sources, localized areas within the illuminated range can be masked by controlling the illumination state or brightness of different high beam channels. The correspondence between target perception and illumination channels, as well as the masking method for different parts of the target area, affects the balance between glare control and road illumination.

[0003] One type of zoned LED matrix masking scheme uses a camera to detect the target and sends information such as the left and right boundary angles of the target's rectangular bounding box to the lighting controller via an onboard communication bus. The lighting controller uses pre-calibrated angle correspondences to determine the LED groups within the target angle range and controls the relevant LED groups to turn off. This method defines the masking range using a rectangular bounding box, and the controlled object is the corresponding LED zone. Another type of high-resolution pixel-level scheme uses a micro-LED array or digital micromirror devices, utilizing a graphics processor or high-performance on-chip system for pixel brightness rendering to achieve more detailed lighting area control. The two schemes differ in the light source resolution and the computing resources of the lighting controller, and also in how they utilize the perception results.

[0004] Rectangular bounding boxes are insufficient to directly reflect the anti-glare requirements of different semantic regions within a target. When applying the same full dark occlusion to the entire area within the bounding box, windows, the vehicle body, and adjacent non-target angles are easily treated together. Target windows require strong occlusion, while maintaining a certain level of brightness for the vehicle body is beneficial for displaying its outline, and road angles excluding the target need to retain illumination. A single occlusion level is not conducive to accommodating these different needs and may cause road areas that do not require occlusion to lose illumination. High-resolution pixel-level solutions can provide more detailed control, but placing pixel-by-pixel brightness rendering on the lighting controller increases the computational resources and hardware costs required by the controller. Therefore, a suitable correspondence between the utilization of target semantic information and the control of lighting channels needs to be established for the division of labor. Summary of the Invention

[0005] This invention provides an adaptive high-beam anti-glare method and system to improve illumination loss in the non-target angle range when performing single-level occlusion based on the target rectangular bounding box, and to reduce the computational burden required for downstream lighting controllers to process images and determine channel brightness.

[0006] To achieve the above objectives, a first aspect of the present invention provides an adaptive high-beam anti-glare method, comprising: The upstream perception controller performs semantic segmentation on the front view image captured by the front view camera to obtain the category label of each pixel; The upstream perception controller divides the horizontal angle range corresponding to the front view image into multiple angle segments according to the angle resolution, obtains the semantic category distribution of the pixel column corresponding to each angle segment, determines the occlusion depth of each angle segment according to the semantic categories and their priorities in the semantic category distribution, and projects the two-dimensional pixel category label along the horizontal angle direction into a one-dimensional angle coding sequence in which each code point corresponds to an angle segment and each code point records the occlusion depth of the corresponding angle segment. The upstream sensing controller sends the one-dimensional angle encoding sequence to the downstream lighting controller via the vehicle communication bus; The downstream lighting controller determines the corresponding code position for the high beam lighting channel whose illumination angle is within the coverage range of the one-dimensional angle encoding sequence, based on the channel's pre-calibrated angle, the starting angle of the one-dimensional angle encoding sequence, and the angle resolution. It then reads the occlusion depth of the code position and determines the target brightness of the channel based on the mapping relationship between occlusion depth and brightness. The downstream lighting controller controls the matrix LED high beams of the corresponding high beam lighting channel to emit light according to the target brightness.

[0007] In one embodiment of the present invention, determining the occlusion depth of each angle segment according to the priority of semantic categories includes: angle segments containing target eye pixels or target window pixels correspond to full occlusion; angle segments containing target eye pixels and target window pixels but containing target vehicle pixels correspond to vehicle body occlusion; angle segments containing target eye pixels, target window pixels, and target vehicle pixels but containing road surface pixels or sky pixels correspond to no occlusion; the remaining angle segments correspond to background occlusion; wherein, the target eye and the target window have the same priority and are both higher than the target vehicle, the target vehicle has a higher priority than the road surface and the sky, and the road surface and the sky have a higher priority than the background; the target brightness of the background occlusion is higher than the target brightness of the vehicle body occlusion and lower than the target brightness of the no occlusion, the target brightness of the vehicle body occlusion is greater than zero, and the target brightness of the full occlusion is zero.

[0008] In one embodiment of the present invention, the one-dimensional angle encoding sequence adopts a four-level occlusion depth encoding, where a depth value of 0 indicates no occlusion, a depth value of 1 indicates background occlusion, a depth value of 2 indicates vehicle body occlusion, and a depth value of 3 indicates full occlusion.

[0009] In one embodiment of the present invention, the upstream perception controller establishes a mapping relationship between pixel columns and horizontal angles based on the horizontal field of view calibrated by the forward-view camera and the width of the forward-view image, uses the mapping relationship to determine the angle segment to which each pixel column belongs, and aggregates the semantic category labels of the corresponding pixel columns according to the angle resolution.

[0010] In one embodiment of the present invention, the pre-calibration angle of each high beam lighting channel is the horizontal angle of the corresponding channel obtained in the lamp calibration process or the vehicle calibration process, and the pre-calibration angle of each channel is written into the memory of the downstream lamp controller as a decoding reference for determining the corresponding code position.

[0011] In one embodiment of the present invention, the horizontal angle range covered by the one-dimensional angle encoding sequence is determined based on the horizontal angle boundary of the target in the front view image and the margin on both sides of the horizontal angle boundary, and is uniformly divided into the plurality of angle segments according to the angle resolution.

[0012] In one embodiment of the present invention, when determining the corresponding code position, the difference between the pre-calibrated angle of the channel and the starting angle is calculated, and the ratio of the difference to the angle resolution is used as the basis for determining the code position index, and the occlusion depth is read according to the code position index.

[0013] In one embodiment of the present invention, when the message is valid and the upstream sensing is valid, the downstream lighting controller further determines the target brightness of the high beam lighting channel whose illumination angle is outside the coverage range of the one-dimensional angle encoding sequence as full brightness.

[0014] In one embodiment of the present invention, when there are multiple targets, the upstream perception controller generates a one-dimensional angle encoding sequence for each target, takes the maximum value of the depth value corresponding to the same horizontal angle segment in the one-dimensional angle encoding sequence of each target, generates a merged one-dimensional angle encoding sequence, and sends the merged one-dimensional angle encoding sequence to the downstream lighting controller through the vehicle communication bus.

[0015] In one embodiment of the present invention, when the corresponding high beam lighting channel is controlled to emit light according to the target brightness, the downstream lighting controller smoothly transitions each channel from the current brightness to the corresponding target brightness through gradual dimming, and drives the matrix LED high beam through the LED driver chip.

[0016] In one embodiment of the present invention, when the message carrying the one-dimensional angle encoding sequence times out, the message verification fails, or the upstream sensing fails, the downstream lighting controller will back down the corresponding high beam lighting channel to a preset safe brightness output.

[0017] In one embodiment of the present invention, the angle resolution is a configurable parameter, and the number of code points in the one-dimensional angle encoding sequence is determined according to the horizontal angle range it covers and the configured angle resolution; the upstream sensing controller generates the one-dimensional angle encoding sequence and the downstream lighting controller determines the corresponding code point using the same sequence starting angle, angle resolution and number of code points.

[0018] In one embodiment of the present invention, the one-dimensional angle encoding sequence adopts N-level occlusion depth encoding, where N is an integer not less than 3; each level of occlusion depth corresponds to a different target brightness, and the stronger the occlusion, the lower the target brightness.

[0019] In a second aspect, the present invention provides an adaptive high-beam anti-glare system, including a forward-looking camera, an upstream sensing controller, a downstream lighting controller, a vehicle communication bus, and matrix LED high beams; The forward-facing camera is used to capture forward-facing images; The upstream perception controller is used to perform semantic segmentation on the front view image, obtain the category label of each pixel, divide the horizontal angle range corresponding to the front view image into multiple angle segments according to the angle resolution, obtain the semantic category distribution of the pixel column corresponding to each angle segment, determine the occlusion depth of each angle segment according to the semantic categories and their priorities in the semantic category distribution, and project the two-dimensional pixel category label along the horizontal angle direction into a one-dimensional angle coding sequence in which each code point corresponds to an angle segment and each code point records the occlusion depth of the corresponding angle segment. The vehicle communication bus is used to transmit the one-dimensional angle encoding sequence generated by the upstream sensing controller to the downstream lighting controller. The downstream lighting controller is used to determine the corresponding code position for the high beam lighting channel whose illumination angle is within the coverage range of the one-dimensional angle encoding sequence, based on the pre-calibrated angle of the channel, the starting angle of the one-dimensional angle encoding sequence and the angle resolution, read the occlusion depth of the code position, determine the target brightness of the channel based on the mapping relationship between occlusion depth and brightness, and control the matrix LED high beam of the corresponding high beam lighting channel to emit light according to the target brightness.

[0020] In one embodiment of the present invention, the upstream perception controller determines the occlusion depth of each angle segment according to the priority of semantic categories, including: angle segments with target eye pixels or target window pixels corresponding to full occlusion; angle segments without target eye pixels or target window pixels but with target vehicle body pixels corresponding to vehicle body occlusion; angle segments without target eye pixels, target window pixels, or target vehicle body pixels but with road surface pixels or sky pixels corresponding to no occlusion; and the remaining angle segments corresponding to background occlusion; wherein, the target eye and the target window have the same priority and are both higher than the target vehicle body, the target vehicle body has a higher priority than the road surface and the sky, and the road surface and the sky have a higher priority than the background; the target brightness of the background occlusion is higher than the target brightness of the vehicle body occlusion and lower than the target brightness of the no occlusion, the target brightness of the vehicle body occlusion is greater than zero, and the target brightness of the full occlusion is zero.

[0021] In a third aspect of the invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the method as described in any of the preceding claims.

[0022] The present invention has the following beneficial effects: This invention projects the two-dimensional semantic segmentation result along the horizontal angle direction into a one-dimensional angle coding sequence, preserving the semantic occlusion requirements of the corresponding angle segment with the occlusion depth of each code point. This sequence serves as the transmission representation between the upstream perception controller and the downstream lighting controller. The downstream determines the code point based on the pre-calibrated angle of the channel and looks up the mapping between depth and brightness, eliminating the need for semantic segmentation or pixel-by-pixel brightness rendering downstream, thus reducing the computational burden on the downstream. In specific implementation, the target eye and the target window share the highest priority, preventing identified eye pixels from being assigned unoccluded output because the same segment contains road or sky pixels; the angle segment corresponding to the vehicle body retains lower brightness, while the angle segment containing road pixels but excluding target eye, target window, and target vehicle body pixels retains illumination. When multiple targets correspond to the same angle segment, a larger occlusion depth is applied to retain stronger anti-glare control. The above control is based on the angle segment and its corresponding lighting channel, and different vertical positions within the same horizontal angle segment are not set with separate brightness settings. Attached Figure Description

[0023] Figure 1 This is the overall flowchart of the adaptive high-beam anti-glare method of the present invention.

[0024] Figure 2 This is a schematic diagram illustrating the architecture and processing division of the adaptive high-beam anti-glare system of the present invention.

[0025] Figure 3 This is a processing logic diagram for determining the occlusion depth according to semantic category priority in this invention.

[0026] Figure 4 This is a schematic diagram comparing the binary occlusion of a rectangular frame with the differentiated occlusion of this invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention.

[0028] In a first embodiment of the present invention, an adaptive high-beam anti-glare method is provided, applied to an upstream perception controller and a downstream lighting controller connected via an onboard communication bus. The upstream perception controller receives images captured by a forward-looking camera and performs image analysis, while the downstream lighting controller controls each high-beam illumination channel of the matrix LED high beam headlights. See also Figure 1 The method includes S1 to S5. Figure 2 The processing boxes in the diagram illustrate the division of labor within the controller, with the calibration benchmark and failure degradation handling corresponding to the channel decoding and output stages; S1 to S5 of the overall method flow are unified under the... Figure 1 Please refer to the following instructions.

[0029] S1. The upstream perception controller performs semantic segmentation on the front view image captured by the front view camera to obtain the category label of each pixel.

[0030] The front view image is an image of the scene in front of the vehicle captured by a camera. The upstream perception controller uses a semantic segmentation network to reason about the image and outputs corresponding category labels for each pixel. Categories related to the occlusion processing in this embodiment include target eye, target window, target vehicle body, road surface, sky, and background. The pixel category labels serve as input for subsequent angular segment category distribution determination.

[0031] In this preferred embodiment, the semantic segmentation network runs on the graphics processor or neural network processor of the upstream perception controller, while the downstream lighting controller uses a microcontroller to perform angle code decoding and brightness control. The upstream component handles image semantic analysis, and the downstream component uses its output one-dimensional angle code sequence to determine the channel brightness.

[0032] In a single-target example, the forward-facing camera captures a 1280×720 pixel image, and the upstream perception controller identifies an oncoming vehicle target in the image. The vehicle's longitudinal axis is taken as the 0° direction of the horizontal angle, and the target's position in the image corresponds to a certain range of horizontal angles.

[0033] S2. The upstream perception controller divides the horizontal angle range corresponding to the front view image into multiple angle segments according to the angle resolution, obtains the semantic category distribution of the pixel column corresponding to each angle segment, determines the occlusion depth of each angle segment according to the semantic categories and their priorities in the semantic category distribution, and projects the two-dimensional pixel category label along the horizontal angle direction into a one-dimensional angle coding sequence in which each code point corresponds to an angle segment and each code point records the occlusion depth of the corresponding angle segment.

[0034] For each angle segment, the semantic category distribution in the corresponding pixel column is obtained, and an occlusion depth is determined based on whether there is a high-priority target category among these pixels. Semantic distillation here involves projecting the two-dimensional pixel classification result onto a one-dimensional horizontal angle code, with each code point storing the occlusion depth of the corresponding angle segment. Arranging these depth values ​​along the angle direction yields the one-dimensional angle code sequence.

[0035] In this preferred embodiment, the upstream perception controller establishes a mapping relationship between pixel columns and horizontal angles based on the horizontal field of view calibrated by the forward-looking camera and the width of the forward-looking image. This mapping relationship is used to determine the angle segment to which each pixel column belongs, and the semantic category labels of the corresponding pixel columns are aggregated according to the angle resolution. The horizontal angle corresponding to the pixel column is based on the camera calibration relationship; the image width and horizontal field of view are used to determine the image range and angle range of this relationship.

[0036] See Figure 3 The target eye and target window are both classified as the highest priority category. Angle segments containing either the target eye pixel or the target window pixel correspond to full occlusion; angle segments without either the target eye pixel or the target window pixel, but containing the target vehicle body pixel, correspond to vehicle body occlusion; angle segments without either the target eye pixel, the target window pixel, or the target vehicle body pixel, but containing the road surface pixel or the sky pixel, correspond to no occlusion; all other angle segments correspond to background occlusion. The target eye and the target window have the same priority and are both higher than the target vehicle body. The target vehicle body has a higher priority than the road surface and the sky, and the road surface and the sky have a higher priority than the background. The target brightness under background occlusion is higher than the target brightness under vehicle body occlusion but lower than the target brightness under no occlusion. The target brightness under vehicle body occlusion is greater than zero, and the target brightness under full occlusion is zero.

[0037] In this preferred embodiment, the one-dimensional angle encoding sequence adopts four-level occlusion depth encoding, where a depth value of 0 indicates no occlusion, a depth value of 1 indicates background occlusion, a depth value of 2 indicates vehicle body occlusion, and a depth value of 3 indicates full occlusion.

[0038] In implementations employing other levels of occlusion depth, this one-dimensional angle encoding sequence uses N-level occlusion depth encoding, where N is an integer not less than 3. Each level of occlusion depth corresponds to a different target brightness, with stronger occlusion depths corresponding to lower target brightness. Each code point still records an occlusion depth for the corresponding angle segment, and the downstream uses a depth-brightness mapping consistent with this N-level encoding. Four-level encoding is one preferred implementation.

[0039] Level 4 coding is used to express the occlusion level of a corresponding angle segment in a single code point. Among them, the target brightness of background occlusion is between that of no occlusion and vehicle occlusion. Vehicle occlusion is used to preserve the vehicle outline illumination, while full occlusion is used to reduce the high-light illumination received by the target's eyes or the target's windows.

[0040] Furthermore, the mapping relationship between occlusion depth and brightness is as follows: the target brightness corresponding to depth values ​​of 0, 1, 2 and 3 is 100%, 60%, 20% and 0% of the full brightness, respectively.

[0041] In this embodiment, the brightness percentage is based on the full brightness of the corresponding channel. Depth value 2 corresponds to 20% low brightness, and depth value 3 corresponds to complete darkness. The brightness mapping is used by the downstream lighting controller in S4.

[0042] In another brightness mapping example, the target brightness for vehicle occlusion is 30% of the corresponding channel's full brightness, background occlusion is 60%, no occlusion is 100%, and full occlusion is 0%. This example replaces the previous mapping with 20% vehicle occlusion; the target brightness for background occlusion is still higher than that for vehicle occlusion but lower than that for no occlusion. Different brightness mappings are configured and used separately.

[0043] In this preferred embodiment, the horizontal angle range covered by the one-dimensional angle encoding sequence is determined based on the horizontal angle boundary of the target in the front view image and the margin on both sides of the horizontal angle boundary, and is uniformly divided into multiple angle segments according to the angle resolution.

[0044] The aforementioned range includes both the angular range corresponding to the target itself and the angular range within a certain margin on both sides of the target, enabling the semantic categories of the target and its surrounding area to participate in encoding. The angular resolution can be configured according to the camera's perception capabilities. In this example, an angular resolution of 0.1° is used, with -6.5° as the starting angle of the encoding sequence. This resolution represents the granularity of the encoding division; the actual spatial control range of the lighting channel is determined by the channel configuration of the matrix LED high beam.

[0045] In this preferred embodiment, the angular resolution is a configurable parameter, and the number of code points in the one-dimensional angular encoding sequence is determined based on the horizontal angular range it covers and the configured angular resolution. The upstream sensing controller generates the one-dimensional angular encoding sequence, and the downstream lighting controller determines the corresponding code point using the same sequence starting angle, angular resolution, and number of code points. When the angular resolution or encoding coverage range is changed, the number of code points in the sequence is adjusted accordingly, so that one code point still corresponds to one angular segment; 0.1° is the preferred angular resolution in this example.

[0046] As shown in Table 1, the angle segment corresponding to -5.3° contains sky and vehicle body pixels but no target eye or window pixels, and is encoded as a depth value of 2; the angle segment corresponding to -5.2° contains vehicle body pixels and is encoded as a depth value of 2; the angle segments corresponding to -4.8° and -4.7° contain window pixels and are both encoded as a depth value of 3; the angle segment corresponding to -3.5° contains vehicle body pixels but no target eye or window pixels and is encoded as a depth value of 2; the angle segments corresponding to -2.2° and -2.1° only contain road surface pixels and are both encoded as a depth value of 0.

[0047]

[0048] Table 1

[0049] Figure 4 This method illustrates the differences in occlusion levels at different horizontal angles: angle segments containing the target eye or window are completely occluded; angle segments containing the vehicle body but not the target eye or window are at lower brightness; and angle segments containing the road surface but not the target eye, window, or vehicle body are fully illuminated. When the same angle segment contains both target eye or window pixels and road surface pixels, it is determined as full occlusion according to the highest priority rule, and the corresponding channel is controlled as a whole. This one-dimensional angle encoding does not set brightness separately for different vertical positions within the same angle segment.

[0050] S3. The upstream sensing controller sends the one-dimensional angle encoding sequence to the downstream lighting controller via the vehicle communication bus.

[0051] In this preferred embodiment, the vehicle communication bus adopts a Controller Area Network (CAN) bus. The transmitted object is a one-dimensional angle encoding sequence generated by S2, and the downstream lighting controller executes S4 after receiving the sequence. The starting angle and angle resolution of the sequence are used to interpret the horizontal angle position corresponding to each code point.

[0052] The upstream sensing controller performs image semantic classification and projection, while the downstream lighting controller uses the projection results for channel control. This division of labor eliminates the need for the downstream to receive and process the complete two-dimensional semantic classification image.

[0053] S4. The downstream lighting controller determines the corresponding code position for the high beam lighting channel whose illumination angle is within the coverage range of the one-dimensional angle encoding sequence, based on the channel's pre-calibrated angle, the starting angle of the one-dimensional angle encoding sequence, and the angle resolution. It reads the occlusion depth of the code position and determines the target brightness of the channel based on the mapping relationship between occlusion depth and brightness.

[0054] Each high beam lighting channel has a pre-calibrated angle used to determine the corresponding one-dimensional angle code position. The downstream lighting controller obtains the depth value from the corresponding code position and then uses the mapping relationship between depth and brightness to determine the target brightness of the channel. In the implementation using four-level encoding, reading a depth value of 2 yields a target brightness of 20%, reading a depth value of 3 yields a target brightness of 0%, and reading a depth value of 0 yields a target brightness of 100%.

[0055] In this preferred embodiment, the pre-calibration angle of each high beam lighting channel is the horizontal angle of the corresponding channel obtained in the lamp calibration process or the vehicle calibration process. The pre-calibration angle of each channel is written into the memory of the downstream lamp controller as a decoding reference for determining the corresponding code bit. The camera pixel column mapping and the channel pre-calibration angle adopt the same horizontal angle reference direction. In this embodiment, the longitudinal axis direction of the vehicle is taken as 0°.

[0056] In this preferred embodiment, when determining the corresponding code position, the difference between the pre-calibrated angle and the starting angle of the channel is calculated, and the ratio of this difference to the angle resolution is used as the basis for determining the code position index. The occlusion depth is then read according to the code position index.

[0057] by Indicates the channel pre-calibration angle, in order to Indicates the starting angle of the encoded sequence, in This represents the angular resolution, where the ratio of the difference to the resolution is... In the example below where the angle and angle segment correspond to the same position, the ratio is an integer, directly representing the index counting from the start code position. Index 0 corresponds to the starting angle position of the sequence.

[0058] For example, channel 8 has a pre-calibration angle of -5.2°, a starting angle of -6.5°, and an angular resolution of 0.1°, corresponding to a ratio of [-5.2 - (-6.5)] / 0.1 = 13. The code depth is 2, and the target brightness is 20%. Channel 9 has a pre-calibration angle of -4.8°, corresponding to a ratio of [-4.8 - (-6.5)] / 0.1 = 17, corresponding to a code depth of 3, and a target brightness of 0%.

[0059] As shown in Table 2, the pre-calibration angle of channel 15 is -2.4°, corresponding to index 41; the pre-calibration angle of channel 16 is -2.0°, corresponding to index 45. The depth value of the corresponding code point for both channels is 0, and the target brightness is 100%. These channel examples are used to illustrate how the obtained depth values ​​are converted into target brightness.

[0060]

[0061] Table 2

[0062] In this preferred embodiment, when the message is valid and the upstream sensing is valid, the downstream lighting controller also determines the target brightness of the high beam lighting channel whose illumination angle is outside the coverage of the one-dimensional angle encoding sequence as full brightness.

[0063] For example, channels with pre-calibrated angles of -20.0°, -18.0°, or +18.0° that are outside the range of the aforementioned encoding sequence do not read depth values ​​from the sequence, and the target brightness remains at 100%. For a luminaire containing 32 high-beam illumination channels, the downstream controller traverses each channel, performing angle range determination, applicable code point reading, and brightness mapping respectively. For a channel within the encoding sequence range, its target brightness is obtained by a single brightness lookup table corresponding to the depth value. This channel decoding process increases linearly with the number of channels, eliminating the need for pixel-by-pixel rendering.

[0064] In this preferred embodiment, when the message carrying the one-dimensional angle encoding sequence times out, message verification fails, or upstream sensing fails, the downstream lighting controller will revert the corresponding high beam lighting channel to a preset safe brightness output. This failure degradation processing applies to high beam lighting channels both within and outside the encoding coverage area; when in a failed state, full-brightness control of channels outside the range is not executed based on abnormal or failed encoding results. The preset safe brightness is the degraded output brightness pre-configured by the lighting controller.

[0065] S5. The downstream lighting controller controls the matrix LED high beams of the corresponding high beam lighting channel to emit light according to the target brightness.

[0066] The downstream lighting controller uses the target brightness determined by S4 to control the illumination of the corresponding channels. The channel corresponding to the window angle segment coded with depth value 3 uses 0% target brightness, the channel corresponding to the vehicle body angle segment coded with depth value 2 uses 20% target brightness, and the channel corresponding to the angle segment coded with depth value 0 uses 100% target brightness.

[0067] In this preferred embodiment, when the corresponding high beam lighting channel is controlled to emit light according to the target brightness, the downstream lighting controller uses gradual dimming to smoothly transition each channel from the current brightness to the corresponding target brightness, and drives the matrix LED high beam lights through the LED driver chip.

[0068] Gradual dimming applies to the transition between the current brightness and the target brightness of each channel. The channel corresponding to the target eye or vehicle window gradually transitions to complete darkness, the channel corresponding to the vehicle body transitions to a lower brightness, and the channel corresponding to the unobstructed angle transitions to or remains fully bright. Gradual dimming does not change the semantic priority of S2 or the target brightness determined by S4.

[0069] In a second embodiment of the present invention, an adaptive high-beam anti-glare method for multi-target scenarios is provided. This method uses image semantic classification, angle encoding, and channel brightness control from the first embodiment, and performs multi-target encoding merging before sending the encoding sequence to the downstream lighting controller.

[0070] When there are multiple targets, the upstream perception controller generates a one-dimensional angle encoding sequence for each target, takes the maximum value of the depth value corresponding to the same horizontal angle segment in the one-dimensional angle encoding sequence of each target, generates a merged one-dimensional angle encoding sequence, and sends the merged one-dimensional angle encoding sequence to the downstream lighting controller through the vehicle communication bus.

[0071] This embodiment employs the four-level depth coding of the first embodiment, where a larger value indicates a stronger degree of occlusion. Therefore, when multiple targets require occlusion within the same horizontal angle segment, the maximum value is taken bit by bit to ensure that the angle segment retains a strong occlusion requirement. This merging process applies to the depth values ​​corresponding to the same angle segment, and the merged result remains a one-dimensional angle coding sequence. Downstream, decoding and output are performed according to S4 and S5 of the first embodiment.

[0072] For example, on a curve at night, an oncoming vehicle is located in an angle range of approximately -4° to -1°, while a bicycle on the right ahead in the same direction is located in an angle range of approximately +1° to +3°. The upstream system generates angle codes for each target separately, and then merges the depth values ​​corresponding to the same angle segment. The angle segment corresponding to the oncoming vehicle's window is fully occluded; the angle segment between two targets containing only road surface pixels and not the target pixels to be occluded is unoccluded.

[0073] For pedestrians or cyclists, after semantic segmentation obtains the target eye category label, both the angle segments containing the target eye pixels and the angle segments containing the target vehicle window pixels are assigned a depth value of 3. Even if the segment also contains road or sky pixels, full occlusion is prioritized. If this highest priority condition is not met, the analysis continues for the target vehicle body, road surface, or sky; otherwise, background occlusion is used. This rule is based on obtaining the corresponding target eye category label.

[0074] For a single target, the encoded sequence of that target can be transmitted directly; for multiple targets, the merged sequence is transmitted. Merging multiple targets does not require additional image analysis steps downstream; the downstream system still determines the target brightness based on the received sequence and channel angle relationships.

[0075] In a third embodiment of the present invention, an adaptive high-beam anti-glare system is provided. See also Figure 2 The system achieves adaptive high-beam anti-glare processing in the first or second embodiment by dividing the processing tasks between the upstream sensing controller and the downstream lighting controller.

[0076] The system includes a forward-facing camera, an upstream sensing controller, a downstream lighting controller, an onboard communication bus, and matrix LED high beams. The forward-facing camera is used to capture forward-facing images.

[0077] The upstream perception controller is used to perform semantic segmentation on the front view image, obtain the category label of each pixel, divide the horizontal angle range corresponding to the front view image into multiple angle segments according to the angle resolution, obtain the semantic category distribution of the pixel column corresponding to each angle segment, determine the occlusion depth of each angle segment according to the semantic categories and their priorities in the semantic category distribution, and project the two-dimensional pixel category label along the horizontal angle direction into a one-dimensional angle coding sequence in which each code point corresponds to an angle segment and each code point records the occlusion depth of the corresponding angle segment. In this preferred embodiment, when the upstream perception controller determines the occlusion depth based on semantic category priority, the angle segment containing target eye pixels or target window pixels corresponds to full occlusion; the angle segment containing target vehicle pixels but not target eye pixels or target window pixels corresponds to vehicle body occlusion; the angle segment containing road surface pixels or sky pixels but not target eye pixels, target window pixels, or target vehicle body pixels corresponds to no occlusion; the remaining angle segments correspond to background occlusion; wherein, the target eye and target window have the same priority and are both higher than the target vehicle body, the target vehicle body has a higher priority than the road surface and sky, and the road surface and sky have a higher priority than the background; the target brightness of the background occluded is higher than the target brightness of the vehicle body occluded but lower than the target brightness of the unoccluded, the target brightness of the vehicle body occluded is greater than zero, and the target brightness of the full occlusion is zero.

[0078] The vehicle communication bus is used to transmit the one-dimensional angle encoding sequence generated by the upstream sensing controller to the downstream lighting controller.

[0079] The downstream lighting controller is used to determine the corresponding code position for the high beam lighting channel whose illumination angle is within the coverage range of the one-dimensional angle coding sequence, based on the pre-calibrated angle of the channel, the starting angle of the one-dimensional angle coding sequence and the angle resolution. It reads the occlusion depth of the code position, determines the target brightness of the channel based on the mapping relationship between occlusion depth and brightness, and controls the matrix LED high beam of the corresponding high beam lighting channel to emit light based on the target brightness.

[0080] The encoded sequence output by the upstream sensing controller serves as the input to the downstream lighting controller, and the semantic classification result participates in lighting control through occlusion depth. The downstream controller reads the corresponding depth value according to the pre-calibrated angle of the channel and uses the obtained target brightness for lighting channel control, thus establishing a correspondence between the image analysis output and the high beam lighting output.

[0081] In this preferred embodiment, the upstream perception controller is configured with a graphics processor or neural network processor to perform semantic segmentation network inference, and the downstream lighting controller uses a microcontroller. The controllers transmit angle-encoded sequences via a CAN bus, and the downstream outputs brightness control for the corresponding channels through an LED driver chip. Encoding, brightness mapping, multi-target merging, and gradual dimming are implemented using the corresponding methods in the first and second embodiments.

[0082] The storage of channel pre-calibrated angles, the mapping of camera pixel columns to horizontal angles, the configurable angle resolution, and the N-level occlusion depth encoding are implemented using the corresponding methods in the first embodiment. When a message times out, message verification fails, or upstream sensing fails, the downstream lighting controller executes the aforementioned preset safety brightness degradation output.

[0083] In a fourth embodiment of the present invention, an electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of any of the methods described above.

[0084] This embodiment also provides a computer-readable storage medium for storing a computer program for adaptive high-beam anti-glare processing.

[0085] The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the adaptive high-beam anti-glare method in the first or second embodiment.

[0086] The computer program executes according to the processing division between the upstream sensing controller and the downstream lighting controller: the upstream side performs image semantic segmentation, determines the angular segment occlusion depth, and generates and transmits the encoded sequence; the downstream side performs encoded sequence reception, channel code bit mapping, target brightness determination, and lighting control. When multiple targets are used, the upstream side performs depth value merging before transmission. The data processed by each program is consistent with the data in the corresponding method steps.

[0087] The present invention has the following beneficial effects: This invention projects the two-dimensional semantic segmentation result along the horizontal angle direction into a one-dimensional angle coding sequence, preserving the semantic occlusion requirements of the corresponding angle segment with the occlusion depth of each code point. This sequence serves as the transmission representation between the upstream perception controller and the downstream lighting controller. The downstream determines the code point based on the pre-calibrated angle of the channel and looks up the mapping between depth and brightness, eliminating the need for semantic segmentation or pixel-by-pixel brightness rendering downstream, thus reducing the computational burden on the downstream. In specific implementation, the target eye and the target window share the highest priority, preventing identified eye pixels from being assigned unoccluded output because the same segment contains road or sky pixels; the angle segment corresponding to the vehicle body retains lower brightness, while the angle segment containing road pixels but excluding target eye, target window, and target vehicle body pixels retains illumination. When multiple targets correspond to the same angle segment, a larger occlusion depth is applied to retain stronger anti-glare control. The above control is based on the angle segment and its corresponding lighting channel, and different vertical positions within the same horizontal angle segment are not set with separate brightness settings.

[0088] The embodiments described above are merely further illustrations of the present invention and are not intended to limit the present invention in any other way. The present invention may have many other embodiments. Without departing from the spirit and essence of the present invention, those skilled in the art can make various corresponding modifications and changes based on the present invention, but all such modifications and changes should fall within the protection scope of the present invention.

Claims

1. An adaptive high-beam anti-glare method, characterized in that, include: The upstream perception controller performs semantic segmentation on the front view image captured by the front view camera to obtain the category label of each pixel; The upstream perception controller divides the horizontal angle range corresponding to the front view image into multiple angle segments according to the angle resolution, obtains the semantic category distribution of the pixel column corresponding to each angle segment, determines the occlusion depth of each angle segment according to the semantic categories and their priorities in the semantic category distribution, and projects the two-dimensional pixel category label along the horizontal angle direction into a one-dimensional angle coding sequence in which each code point corresponds to an angle segment and each code point records the occlusion depth of the corresponding angle segment. The upstream sensing controller sends the one-dimensional angle encoding sequence to the downstream lighting controller via the vehicle communication bus; The downstream lighting controller determines the corresponding code position for the high beam lighting channel whose illumination angle is within the coverage range of the one-dimensional angle encoding sequence, based on the channel's pre-calibrated angle, the starting angle of the one-dimensional angle encoding sequence, and the angle resolution. It then reads the occlusion depth of the code position and determines the target brightness of the channel based on the mapping relationship between occlusion depth and brightness. The downstream lighting controller controls the matrix LED high beams of the corresponding high beam lighting channel to emit light according to the target brightness.

2. The adaptive high-beam anti-glare method according to claim 1, characterized in that, The determination of the occlusion depth of each angle segment based on the priority of semantic categories includes: angle segments containing target eye pixels or target window pixels correspond to full occlusion; angle segments containing target eye pixels and target window pixels but target vehicle pixels correspond to vehicle body occlusion; angle segments containing target eye pixels, target window pixels, and target vehicle pixels but road surface pixels or sky pixels correspond to no occlusion; the remaining angle segments correspond to background occlusion; wherein, the target eye and the target window have the same priority and are both higher than the target vehicle, the target vehicle has a higher priority than the road surface and the sky, and the road surface and the sky have a higher priority than the background; the target brightness of the background occlusion is higher than the target brightness of the vehicle body occlusion and lower than the target brightness of the no-occlusion, the target brightness of the vehicle body occlusion is greater than zero, and the target brightness of the full occlusion is zero.

3. The adaptive high-beam anti-glare method according to claim 2, characterized in that, The one-dimensional angle encoding sequence adopts a four-level occlusion depth encoding, where a depth value of 0 indicates no occlusion, a depth value of 1 indicates background occlusion, a depth value of 2 indicates vehicle body occlusion, and a depth value of 3 indicates full occlusion.

4. The adaptive high-beam anti-glare method according to claim 1, characterized in that, The upstream perception controller establishes a mapping relationship between pixel columns and horizontal angles based on the horizontal field of view calibrated by the forward-view camera and the width of the forward-view image. It uses the mapping relationship to determine the angle segment to which each pixel column belongs and aggregates the semantic category labels of the corresponding pixel columns according to the angle resolution.

5. The adaptive high-beam anti-glare method according to claim 1, characterized in that, The pre-calibration angle of each high beam lighting channel is the horizontal angle of the corresponding channel obtained in the lamp calibration process or the vehicle calibration process. The pre-calibration angle of each channel is written into the memory of the downstream lamp controller as a decoding reference for determining the corresponding code position.

6. The adaptive high-beam anti-glare method according to claim 1, characterized in that, The horizontal angle range covered by the one-dimensional angle encoding sequence is determined based on the horizontal angle boundary of the target in the front view image and the margin on both sides of the horizontal angle boundary, and is uniformly divided into the multiple angle segments according to the angle resolution.

7. The adaptive high-beam anti-glare method according to claim 1, characterized in that, When determining the corresponding code position, the difference between the pre-calibrated angle of the channel and the starting angle is calculated, and the ratio of the difference to the angle resolution is used as the basis for determining the code position index. The occlusion depth is then read according to the code position index.

8. The adaptive high-beam anti-glare method according to claim 1, characterized in that, When the message is valid and the upstream sensing is valid, the downstream lighting controller also determines the target brightness of the high beam lighting channel whose illumination angle is outside the coverage of the one-dimensional angle encoding sequence as full brightness.

9. The adaptive high-beam anti-glare method according to claim 3, characterized in that, When multiple targets exist, the upstream perception controller generates a one-dimensional angle encoding sequence for each target, takes the maximum value of the depth value corresponding to the same horizontal angle segment in the one-dimensional angle encoding sequence of each target, generates a merged one-dimensional angle encoding sequence, and sends the merged one-dimensional angle encoding sequence to the downstream lighting controller through the vehicle communication bus.

10. The adaptive high-beam anti-glare method according to claim 1, characterized in that, When the corresponding high beam lighting channel is controlled to emit light according to the target brightness, the downstream lighting controller smoothly transitions each channel from the current brightness to the corresponding target brightness through gradual dimming, and drives the matrix LED high beam through the LED driver chip.

11. The adaptive high-beam anti-glare method according to claim 1, characterized in that, When the message carrying the one-dimensional angle encoding sequence times out, the message verification fails, or the upstream sensing fails, the downstream lighting controller will revert the corresponding high beam lighting channel to the preset safe brightness output.

12. The adaptive high-beam anti-glare method according to claim 1, characterized in that, The angle resolution is a configurable parameter, and the number of code points in the one-dimensional angle encoding sequence is determined according to the horizontal angle range it covers and the configured angle resolution. The upstream sensing controller generates the one-dimensional angle encoding sequence and the downstream lighting controller determines the corresponding code point using the same sequence starting angle, angle resolution, and number of code points.

13. The adaptive high-beam anti-glare method according to claim 1, characterized in that, The one-dimensional angle encoding sequence adopts N-level occlusion depth encoding, where N is an integer not less than 3; each level of occlusion depth corresponds to a different target brightness, and the stronger the occlusion, the lower the target brightness.

14. An adaptive high-beam anti-glare system, characterized in that, This includes a forward-facing camera, an upstream sensing controller, a downstream lighting controller, an in-vehicle communication bus, and matrix LED high beams; The forward-facing camera is used to capture forward-facing images; The upstream perception controller is used to perform semantic segmentation on the front view image, obtain the category label of each pixel, divide the horizontal angle range corresponding to the front view image into multiple angle segments according to the angle resolution, obtain the semantic category distribution of the pixel column corresponding to each angle segment, determine the occlusion depth of each angle segment according to the semantic categories and their priorities in the semantic category distribution, and project the two-dimensional pixel category label along the horizontal angle direction into a one-dimensional angle coding sequence in which each code point corresponds to an angle segment and each code point records the occlusion depth of the corresponding angle segment. The vehicle communication bus is used to transmit the one-dimensional angle encoding sequence generated by the upstream sensing controller to the downstream lighting controller. The downstream lighting controller is used to determine the corresponding code position for the high beam lighting channel whose illumination angle is within the coverage range of the one-dimensional angle encoding sequence, based on the pre-calibrated angle of the channel, the starting angle of the one-dimensional angle encoding sequence and the angle resolution, read the occlusion depth of the code position, determine the target brightness of the channel based on the mapping relationship between occlusion depth and brightness, and control the matrix LED high beam of the corresponding high beam lighting channel to emit light according to the target brightness.

15. The adaptive high-beam anti-glare system according to claim 14, characterized in that, The upstream perception controller determines the occlusion depth of each angle segment based on the priority of semantic categories, including: angle segments with target eye pixels or target window pixels correspond to full occlusion; angle segments without target eye pixels or target window pixels but with target vehicle pixels correspond to vehicle body occlusion; angle segments without target eye pixels, target window pixels, or target vehicle pixels but with road surface pixels or sky pixels correspond to no occlusion; the remaining angle segments correspond to background occlusion; wherein, the target eye and the target window have the same priority and are both higher than the target vehicle, the target vehicle has a higher priority than the road surface and the sky, and the road surface and the sky have a higher priority than the background; the target brightness of the background occlusion is higher than the target brightness of the vehicle body occlusion and lower than the target brightness of the no-occlusion, the target brightness of the vehicle body occlusion is greater than zero, and the target brightness of the full occlusion is zero.

16. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 13.