Vehicle lighting control system and method based on sparse gated hybrid expert model

Abstract

This disclosure provides a vehicle lighting control system and method based on a sparse-gated hybrid expert model, relating to the field of vehicle lighting adaptive control technology. The method includes: acquiring and preprocessing ultrasonic, image, and audio multimodal signals; inputting the preprocessed multimodal signals into a sparse-gated hybrid expert model, then sending the ultrasonic, image, and audio signals into independent expert networks. Each expert network is responsible for analyzing a specific type of input signal. The ultrasonic expert network analyzes the ultrasonic signal and outputs the distance and category of the object in front; the image expert network analyzes image features to obtain the distance and category of the object in front and a preliminary vehicle lighting code; the audio expert network analyzes sound features to obtain the distance and category of the object in front; finally, a routing network gating mechanism is used to select the outputs of two expert networks for weighted combination to generate the final vehicle lighting code, thus obtaining the vehicle lighting control signal.

Description

Technical Field

[0001] This disclosure relates to the field of adaptive vehicle lighting control technology, specifically to a vehicle lighting control system and method based on a sparse gated hybrid expert model. Background Technology

[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.

[0003] While traditional car high beams provide a longer field of vision at night, they can also glare at oncoming vehicles, impairing their vision and increasing the risk of traffic accidents. To address this issue, Adaptive High Beam (ADB) systems were developed. ADB systems dynamically adjust the range of the high beams based on road conditions and the surrounding environment to provide sufficient illumination while avoiding glare.

[0004] Existing ADB systems primarily rely on single camera signals to detect objects and road conditions ahead. However, single sensor signals often have limitations, such as low recognition accuracy in adverse weather conditions, insufficient precision in detecting objects ahead, and a tendency to misjudge or miss detections, making it difficult to comprehensively and accurately identify objects in various environments. Existing multi-sensor fusion methods face significant challenges due to substantial differences in data processing methods and algorithms among different sensors. Furthermore, existing ADB systems suffer from poor real-time performance, slow response times, and an inability to adjust lights promptly, impacting driving safety. The computational load for data processing is also substantial, especially with complex image processing algorithms leading to high latency. Performance in complex environments (such as rain, snow, and strong light interference) is unstable, prone to misjudgments, and lacks adaptability to environmental changes, demonstrating a lack of effective robust design. Summary of the Invention

[0005] To address the aforementioned issues, this disclosure proposes a vehicle lighting control system and method based on a sparse gated hybrid expert model (SMoE). By fusing signals from multiple sensors and utilizing a linear layer and Top-K gating mechanism through a router network, the system processes multiple signals separately and combines the individual outputs into a final vehicle lighting code. This enables high-precision perception of the environment ahead and intelligent control of the high beams.

[0006] According to some embodiments, the present disclosure adopts the following technical solutions:

[0007] The vehicle lighting control method based on sparse gated hybrid expert models includes:

[0008] Acquire and preprocess ultrasound, image, and audio multimodal signals;

[0009] The preprocessed multimodal signal is input into a sparse gated hybrid expert model, and the output is the headlight control signal to achieve adaptive adjustment of the headlights;

[0010] The preprocessed multimodal signals are input into a sparsely gated hybrid expert model. Ultrasonic, image, and audio signals are then fed into separate expert networks. Each expert network is responsible for analyzing a specific type of input signal. The ultrasonic expert network analyzes the ultrasonic signal and outputs the distance and category of the object in front. The image expert network analyzes image features to obtain the distance and category of the object in front and a preliminary headlight code. The audio expert network analyzes sound features to obtain the distance and category of the object in front. Finally, the outputs of two expert networks are selected and weighted by a routing network gating mechanism to generate the final headlight code and obtain the headlight control signal.

[0011] According to some embodiments, the present disclosure adopts the following technical solutions:

[0012] The vehicle lighting control system based on a sparse gated hybrid expert model includes:

[0013] The data acquisition module is used to acquire and preprocess ultrasonic, image, and audio multimodal signals.

[0014] The headlight control module is used to input the preprocessed multimodal signal into the sparse gated hybrid expert model and output the headlight control signal to realize the adaptive adjustment of the headlight;

[0015] The preprocessed multimodal signals are input into a sparsely gated hybrid expert model. Ultrasonic, image, and audio signals are then fed into separate expert networks. Each expert network is responsible for analyzing a specific type of input signal. The ultrasonic expert network analyzes the ultrasonic signal and outputs the distance and category of the object in front. The image expert network analyzes image features to obtain the distance and category of the object in front and a preliminary headlight code. The audio expert network analyzes sound features to obtain the distance and category of the object in front. Finally, the outputs of two expert networks are selected and weighted by a routing network gating mechanism to generate the final headlight code and obtain the headlight control signal.

[0016] According to some embodiments, the present disclosure adopts the following technical solutions:

[0017] A computer program product includes a computer program that, when executed by a processor, implements the vehicle lighting control method based on a sparse gating hybrid expert model.

[0018] According to some embodiments, the present disclosure adopts the following technical solutions:

[0019] A non-transitory computer-readable storage medium is provided for storing computer instructions, which, when executed by a processor, implement the aforementioned vehicle lighting control method based on a sparse gating hybrid expert model.

[0020] According to some embodiments, the present disclosure adopts the following technical solutions:

[0021] An electronic device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the vehicle lighting control method based on a sparse gating hybrid expert model.

[0022] Compared with the prior art, the beneficial effects of this disclosure are as follows:

[0023] This disclosure presents a vehicle headlight control method based on a sparse gated hybrid expert model. The adaptive high beam system of this disclosure can accurately detect objects ahead in various environments and intelligently adjust the illumination range and intensity of the headlights, significantly improving the safety and comfort of nighttime driving. This disclosure enhances safety by intelligently controlling the illumination range and intensity of the headlights, reducing interference with other road users and improving nighttime driving safety; it also enhances comfort by adjusting the headlights in real time according to changes in the forward environment, improving the driver's visibility and comfort.

[0024] This disclosure presents a vehicle lighting control method based on a sparse gated hybrid expert model. By fusing signals from multiple sensors, the robustness and accuracy of the system are improved. The method employs a sparse hybrid expert model (SMoE) to analyze the world, increasing the number of model parameters while maintaining low computational cost. By selecting two expert networks to process each input signal, the robustness and accuracy of the system are significantly improved while maintaining low computational cost. Experimental results show that the system of this disclosure can work effectively in various environments, significantly improving the safety and comfort of nighttime driving. Attached Figure Description

[0025] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.

[0026] Figure 1 This is a schematic diagram of the method flow of an embodiment of the present disclosure. Detailed Implementation

[0027] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0028] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0029] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this disclosure. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0030] Example 1

[0031] One embodiment of this disclosure provides a vehicle lighting control method based on a sparse gated hybrid expert model, comprising the following steps:

[0032] Step 1: Acquire and preprocess ultrasound, image, and audio multimodal signals;

[0033] Step 2: Input the preprocessed multimodal signal into the sparse gated hybrid expert model and output the headlight control signal to achieve adaptive adjustment of the headlights;

[0034] The preprocessed multimodal signals are input into a sparsely gated hybrid expert model. Ultrasonic, image, and audio signals are then fed into separate expert networks. Each expert network is responsible for analyzing a specific type of input signal. The ultrasonic expert network analyzes the ultrasonic signal and outputs the distance and category of the object in front. The image expert network analyzes image features to obtain the distance and category of the object in front and a preliminary headlight code. The audio expert network analyzes sound features to obtain the distance and category of the object in front. Finally, the outputs of two expert networks are selected and weighted by a routing network gating mechanism to generate the final headlight code and obtain the headlight control signal.

[0035] As one embodiment, the specific implementation process of the vehicle lighting control method based on a sparse gating hybrid expert model disclosed herein is as follows:

[0036] Step 1: Acquire and preprocess ultrasound, image, and audio multimodal signals;

[0037] Specifically, the ultrasonic signal is acquired using an ultrasonic sensor, which obtains the ultrasonic signal reflected back from the object in front by emitting ultrasonic signals; the image signal is acquired using a camera, and the audio signal is acquired using a microphone.

[0038] The preprocessing method includes signal normalization and noise reduction.

[0039] Step 2: Input the preprocessed multimodal signal into the sparse gated hybrid expert model and output the headlight control signal to realize the adaptive adjustment of the headlight;

[0040] Specifically, the sparse-gated hybrid expert model is based on a transformer architecture and uses a sparse hybrid expert layer. This layer consists of multiple expert network sub-models, each processing a specific type of input signal. Following the sparse hybrid expert layer is a routing network, which is a trainable gating network used to select two expert networks to weight and process the current input signal, outputting a sparse n-dimensional vector. This includes:

[0041] Normalized ultrasonic echo features, image feature matrices, audio, lidar signal features, and infrared signal features are embedded and aligned across modally, and temporal correlations are established through a positional coding layer. A multi-head self-attention mechanism is employed to capture the dynamic interaction relationships between cross-modal features. The preprocessed ultrasonic, image, and audio signals are input into expert networks, with each expert specializing in processing a specific signal. Each expert network contains three layers of gated recurrent units (a trainable gated network based on a transformer architecture, with input being a concatenated vector of multimodal features) and an independent feature extraction branch that calculates the cosine similarity between the input signal and each expert network, retaining the corresponding top experts.

[0042] Furthermore, the results processed by each expert network are input to the routing network. The routing network selects two expert networks to process the current input signal, and outputs the final headlight control signal by weighting. The generated headlight control signal is sent to the headlight control system. The control system adjusts the illumination range and intensity of the headlights according to the headlight control signal to achieve the function of adaptive adjustment of the high beam.

[0043] Specifically, a three-layer gated recurrent unit (a trainable gated network designed based on a transformer architecture, with input being a multimodal feature concatenation vector) is equipped with an independent feature extraction branch to calculate the cosine similarity between the input signal and each expert network, and retain the corresponding top experts.

[0044] KL divergence evaluation was performed on the top experts, and the output weight vectors of the two best experts were selected and normalized by Softmax. The selected two experts processed the input features separately, and then the features were concatenated according to their weights.

[0045] F_fusion=w_1·Expert_1(X)⊕w_2·Expert_2(X)

[0046] Wherein, ⊕ represents the channel dimension splicing operation, which preserves the independence of each expert's output features (by using weighted summation or attention mechanism to fuse the output features of two experts).

[0047] As one embodiment, in the vehicle lighting control method based on a sparse-gated hybrid expert model disclosed herein, the Moe system can process signals from different sensors, including ultrasonic signals, camera signals, and audio signals. The ultrasonic sensor provides distance and category information, the camera provides image information and generates preliminary vehicle lighting codes through a Transformer layer, and the audio signal provides ambient sound information, assisting the Moe system in better outputting the final vehicle lighting codes. The Moe system contains multiple expert networks, each responsible for processing a specific type of input signal. A suitable expert network is selected to process the current input signal through a router network, ensuring that each input signal is processed by the most suitable expert network, thereby improving the system's accuracy and real-time performance. The Moe system employs a sparse gating mechanism, activating only a few expert networks for each input signal, rather than all of them. The Moe system ensures balanced load across all expert networks through a dynamic load balancing mechanism. By introducing a load balancing loss function, the Moe system can effectively allocate computational resources, avoiding overload of certain expert networks and improving the overall system performance. By training the Moe system, it can automatically select the most suitable expert network based on the characteristics of the input signal, thereby improving the system's robustness and adaptability.

[0048] The ultrasonic sensor is a non-contact measurement tool that measures the distance to objects by emitting and receiving ultrasonic signals. The ultrasonic sensor used in this disclosure is trained to output the distance to an object and the object's category, much like a bat's brain. Specifically, the ultrasonic model takes ultrasonic signals as input and outputs the distance to the object and the object's category. A camera captures images of the foreground environment and outputs the distance to the object and the object's category through a trained neural network model. Furthermore, the camera signal, after processing through a Transformer layer, is able to output vehicle light codes.

[0049] The Mixture of Experts (MoE) model is a neural network architecture that processes input signals through a combination of multiple expert networks (i.e., sub-models). Each expert network is responsible for processing a specific type of input signal, and the outputs of two expert networks are selected by a routing network and weighted together to generate the final headlight control signal. This architecture can significantly increase the number of model parameters while maintaining low computational cost. The hybrid expert model disclosed herein is based on the Transformer architecture and uses a Sparse Mixture of Experts (SMoE) layer. The routing network is a trainable gating network used to select two expert networks to process the current input signal. The output of the routing network is a sparse n-dimensional vector, where n is the number of expert networks. By using the sparse mixture of expert models, this disclosure can significantly increase the number of model parameters while maintaining low computational cost. Specifically, each input signal only requires processing two expert networks, so the actual number of parameters used per input signal is approximately 1.3 billion, while the total number of parameters is 4.7 billion. This architecture enables the model to perform well on large-scale datasets while maintaining high computational efficiency. The acquired data is first preprocessed, including normalization and noise reduction, to ensure the quality of the input signal. The preprocessed signal is then fed into the corresponding neural network model for further processing: ultrasonic signals are processed by the ultrasonic model, outputting the distance to the object and the object's category. Camera signals are processed by the camera model, outputting the distance to the object, the object's category, and the vehicle light code.

[0050] As one embodiment, LiDAR signals and infrared signals can also be acquired. The LiDAR signals are processed through a LiDAR model to output the distance to the object and the object's category, used for 3D spatial modeling. The infrared signals are processed through an infrared model to determine the presence of pedestrians. The processed signals are sent to a routing network, which selects two expert networks to process the current input signal. The generated headlight control signals are sent to the headlight control system, which adjusts the headlight illumination range and intensity according to the signals, thereby achieving the function of adaptive high beams.

[0051] The ultrasonic distance measurement disclosed herein is performed by uploading the measurement data to the server after it is completed in the vehicle body domain control. The measurement formula is as follows;

[0052]

[0053] Where d is the distance, c is the speed of sound (343 m / s), and t is the round-trip time of the ultrasonic signal.

[0054] Adaptive Driving Beam (ADB) is designed to improve nighttime driving safety and comfort by intelligently sensing the surrounding environment and automatically adjusting the vehicle's high beams. ADB aims to prevent dazzling oncoming drivers. It automatically detects obstacles and vehicles ahead and adjusts the beam range to ensure optimal illumination without affecting other drivers.

[0055] This disclosure provides an adaptive high beam system, specifically comprising the following components: an ultrasonic sensor for detecting the distance and category of objects ahead. The ultrasonic sensor operates similarly to echolocation in bats, calculating distance by emitting ultrasonic waves and receiving reflected waves. An ultrasonic model is trained to output the distance to the object and the object's category. A camera is used to capture image information ahead. A camera model is trained to output the distance to the object and the object's category, and generates headlight codes through a Transformer layer.

[0056] Ultrasonic sensors are mounted on the left and right sides of the front of the vehicle and connected to the main control unit via a CAN bus. A camera is mounted on the front of the vehicle and connected to the main control unit via a high-speed Ethernet interface. The headlight controller receives control signals from the main control unit and adjusts the illumination range and intensity of the high beams. The ultrasonic sensors emit ultrasonic signals, receive reflected waves, and calculate distances. Simultaneously, the reflected ultrasonic signals are transmitted to the cloud, where an ultrasonic model processes the signals and outputs the distance to objects and the object's category. The camera captures images of the front, and the camera model outputs the distance to objects and the object's category based on the image content. This information is converted into headlight codes through a Transformer layer. The headlight controller adjusts the illumination range and intensity of the high beams based on the headlight codes generated by the main control unit. The training process of the ultrasonic model first requires data acquisition, using ultrasonic sensors to collect ultrasonic signals from the front environment. Data preprocessing involves normalizing and denoising the acquired signals. Model training uses a training dataset to train the ultrasonic model, employing the Adam optimizer with an initial linear learning rate that gradually decreases. Model evaluation assesses the model's performance on the test dataset to ensure accurate output of distance to objects and object categories. Experimental results show that the model learns the different reflections of different objects and utilizes reflection, Doppler effect, and transmission to acquire and process information about the object. The training process for the camera model also begins with data acquisition, capturing images of the surrounding environment through a camera.

[0057] Data preprocessing involves normalizing and denoising the acquired images, then feeding the signals into the Moe model for training. Model evaluation assesses the model's performance on the test dataset to ensure that the model can accurately output the distance to the object, the object's category, and the vehicle light code.

[0058] As one implementation, the mapping strategy for vehicle headlight position encoding can generally be adjusted based on visual information features and target detection location information to establish a mapping layer from the image coordinate system to the vehicle headlight control signal. A binary search algorithm is used for fast spatial localization of the target coordinates. The target detection algorithm outputs in two formats: COCO and VOC. The VOC format allows for direct location of the target's center coordinates using a binary search method after restoring the coordinate information. The four values ​​output by the VOC format (x1, x2, y1, y2) are directly used as ((x1+x2) / 2, (y1+y2) / 2). Because the VOC format outputs normalized values, to locate the center point of the target detection, the original coordinates need to be used. Once the center position of the target is determined, it can be determined whether it is on the left or right side of the vehicle. Since high beam LEDs typically use 8 LEDs, the range can be determined based on the number of LEDs used in different vehicle models. Knowing the center position of the target detection, the binary search algorithm can determine the range where the target's center position is located. The adaptive high beam (ADB) then determines which LEDs need to be turned off based on this range. The output is in COCO format, so y2 = y + h, x2 = x + w. The target center position is determined according to the above scheme, and the algorithm is repeated. Alternatively, reinforcement learning can be used to define the headlight coding layer. However, the accuracy of the headlight coding layer output by reinforcement learning is questionable, and the security of deep learning and reinforcement learning on in-vehicle systems is debatable. Therefore, this patent still uses the traditional manual program to set the headlight coding layer, including motor control, where the output of the neural network is processed by a program before being output.

[0059] Experiments were conducted in various scenarios, including urban roads, highways, and rural roads. The results show that the adaptive high beam system of this invention can accurately detect objects ahead in various environments and intelligently adjust the illumination range and intensity of the headlights, significantly improving the safety and comfort of nighttime driving. This patent enhances safety by intelligently controlling the illumination range and intensity of the headlights, reducing interference with other road users and improving nighttime driving safety. It also improves comfort by adjusting the headlights in real time according to changes in the road environment, enhancing the driver's visibility and comfort. Multimodal fusion improves the robustness and accuracy of the system by fusing signals from multiple sensors. Finally, it achieves computational efficiency by employing a sparse hybrid expert model (SMoE), maintaining low computational cost while increasing the number of model parameters.

[0060] Example 2

[0061] One embodiment of this disclosure provides a vehicle lighting control system based on a sparse gated hybrid expert model, including:

[0062] The data acquisition module is used to acquire and preprocess ultrasonic, image, and audio multimodal signals.

[0063] The headlight control module is used to input the preprocessed multimodal signal into the sparse gated hybrid expert model and output the headlight control signal to realize the adaptive adjustment of the headlight;

[0064] The preprocessed multimodal signals are input into a sparsely gated hybrid expert model. Ultrasonic, image, and audio signals are then fed into separate expert networks. Each expert network is responsible for analyzing a specific type of input signal. The ultrasonic expert network analyzes the ultrasonic signal and outputs the distance and category of the object in front. The image expert network analyzes image features to obtain the distance and category of the object in front and a preliminary headlight code. The audio expert network analyzes sound features to obtain the distance and category of the object in front. Finally, the outputs of two expert networks are selected and weighted by a routing network gating mechanism to generate the final headlight code and obtain the headlight control signal.

[0065] Furthermore, the sparse gated hybrid expert model is based on a transformer architecture and uses a sparse hybrid expert layer. This layer consists of multiple expert network sub-models, each processing a specific type of input signal. Following the sparse hybrid expert layer is a routing network, which is a trainable gated network used to select two expert networks to weightedly process the current input signal, outputting a sparse n-dimensional vector. The outputs of each expert network are input to the routing network, which selects two expert networks to process the current input signal, weighting and outputting the final headlight control signal. This generated headlight control signal is sent to the headlight control system, which adjusts the headlight illumination range and intensity based on the signal, achieving adaptive high beam adjustment.

[0066] Example 3

[0067] One embodiment of this disclosure provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned vehicle lighting control method based on a sparse gating hybrid expert model.

[0068] Example 4

[0069] One embodiment of this disclosure provides a non-transitory computer-readable storage medium for storing computer instructions. When these computer instructions are executed by a processor, they implement the vehicle lighting control method based on a sparse gating hybrid expert model.

[0070] Example 5

[0071] One embodiment of this disclosure provides an electronic device, including a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the vehicle lighting control method based on a sparse gating hybrid expert model.

[0072] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0073] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0074] While the specific embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this disclosure. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this disclosure are still within the scope of protection of this disclosure.

Claims

1. A vehicle lighting control method based on a sparse gated hybrid expert model, characterized in that, include: Acquire and preprocess ultrasound, image, and audio multimodal signals; The preprocessed multimodal signal is input into a sparse gated hybrid expert model, and the output is the headlight control signal to achieve adaptive adjustment of the headlights; The preprocessed multimodal signals are input into a sparsely gated hybrid expert model. Ultrasonic, image, and audio signals are then fed into independent expert networks. Each expert network is responsible for analyzing a specific type of input signal. The ultrasonic expert network analyzes the ultrasonic signal and outputs the distance and category of the object in front. The image expert network analyzes image features to obtain the distance and category of the object in front and a preliminary headlight code. The audio expert network analyzes sound features to obtain the distance and category of the object in front. Finally, the outputs of two expert networks are selected through a routing network gating mechanism and weighted to generate the final headlight code, thus obtaining the headlight control signal. The sparse-gated hybrid expert model is based on a transformer architecture and uses a sparse hybrid expert layer. This layer consists of multiple expert network sub-models, each processing a specific type of input signal. Following the sparse hybrid expert layer is a routing network, which is a trainable gating network used to select two expert networks to weight the current input signal, resulting in a sparse output. n Dimensional vector.

2. The vehicle lighting control method based on a sparse gated hybrid expert model as described in claim 1, characterized in that, The ultrasonic signal is acquired using an ultrasonic sensor, which obtains the ultrasonic signal reflected back from the object in front by emitting ultrasonic signals; the image signal is acquired by a camera, and the audio signal is acquired by a microphone; the preprocessing method is signal normalization and noise reduction.

3. The vehicle lighting control method based on a sparse gated hybrid expert model as described in claim 1, characterized in that, The results processed by each expert network are input to the routing network. The routing network selects two expert networks to process the current input signal, and outputs the final headlight control signal by weighting. The generated headlight control signal is sent to the headlight control system. The control system adjusts the illumination range and intensity of the headlights according to the headlight control signal to achieve the function of adaptive adjustment of the high beam.

4. A vehicle lighting control system based on a sparse gated hybrid expert model, characterized in that, include: The data acquisition module is used to acquire and preprocess ultrasonic, image, and audio multimodal signals. The headlight control module is used to input the preprocessed multimodal signal into the sparse gated hybrid expert model and output the headlight control signal to realize the adaptive adjustment of the headlight; The preprocessed multimodal signals are input into a sparsely gated hybrid expert model. Ultrasonic, image, and audio signals are then fed into independent expert networks. Each expert network is responsible for analyzing a specific type of input signal. The ultrasonic expert network analyzes the ultrasonic signal and outputs the distance and category of the object in front. The image expert network analyzes image features to obtain the distance and category of the object in front and a preliminary headlight code. The audio expert network analyzes sound features to obtain the distance and category of the object in front. Finally, the outputs of two expert networks are selected through a routing network gating mechanism and weighted to generate the final headlight code, thus obtaining the headlight control signal. The sparse-gated hybrid expert model is based on a transformer architecture and uses a sparse hybrid expert layer. This layer consists of multiple expert network sub-models, each processing a specific type of input signal. Following the sparse hybrid expert layer is a routing network, which is a trainable gating network used to select two expert networks to weight the current input signal, resulting in a sparse output. n Dimensional vector.

5. The vehicle lighting control system based on a sparse gated hybrid expert model as described in claim 4, characterized in that, The results processed by each expert network are input to the routing network. The routing network selects two expert networks to process the current input signal, and outputs the final headlight control signal by weighting. The generated headlight control signal is sent to the headlight control system. The control system adjusts the illumination range and intensity of the headlights according to the headlight control signal to achieve the function of adaptive adjustment of the high beam.

6. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the vehicle lighting control method based on a sparse gating hybrid expert model as described in any one of claims 1-3.

7. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement the vehicle lighting control method based on a sparse gating hybrid expert model as described in any one of claims 1-3.

8. An electronic device, characterized in that, include: The device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the vehicle lighting control method based on a sparse gating hybrid expert model as described in any one of claims 1-3.

Images