ChatGPT-based automobile light auxiliary control method and system

Abstract

The invention discloses an automobile light auxiliary control method and system based on ChatGPT, and belongs to the technical field of automobile intelligent driving, and the method comprises the following steps: S1, obtaining external environment data; s2, acquiring a control instruction; and S3, comprehensive judgment: a safer and more effective light use instruction can be obtained by integrating environment information and driver judgment.

Description

Technical Field

[0001] The present invention relates to the technical field of automobile intelligent driving, and in particular to an automobile lighting auxiliary control system based on ChatGPT. Background Art

[0002] As a means of transportation, cars are widely used in modern society. Nowadays, high-end cars are equipped with intelligent systems to enhance the driving experience. The lighting system on the car is indispensable for car driving. The existing car lighting systems on the market are composed of multiple functional modules. According to whether the lighting system is in contact with the external environment, it can be divided into indoor lighting system and outdoor lighting system. The indoor lighting system includes atmosphere lights, reading lights, foot lights, instrument indicator lights, etc. The outdoor lighting system includes matrix LED headlights, adaptive high beams (ADB), adaptive front lights (AFS), digital projection lights (DLP), front (rear) position lights, front (rear) turn signals, rear taillights, reversing lights, brake lights, etc. Whether it is an indoor lighting system or an outdoor lighting system, lighting control requires Different software / hardware drive systems need to be matched separately, and different headlights have different levels of functional safety requirements. In particular, some lights for assisted driving require a lot of computing resources. Different module lighting systems need to make algorithms based on different chips. Therefore, hardware, software, computing power, and control strategies all need to be developed and designed separately. There is stacking in the algorithm-related parts, waste in the space-related parts, and the cost cannot be effectively saved. In terms of headlight control, although the body domain controller BDCU can be used to distribute commands for control, the car headlight system is not intelligent enough in terms of automatic recognition of driving scenarios and automated driving, and cannot understand humans' actual lighting needs. The human-computer interaction performance of the headlight system is not high enough.

[0003] When a bug occurs in the program or an electrical device is interfered with externally, errors may occur, leading to incorrect operation. For example, when the adaptive front lighting (AFS) lighting system passes through a turning area at night, the vehicle lighting system is triggered to automatically reduce the brightness or turn off the lights because it happens to pass through a strong light illumination area when turning. When the car leaves the strong light area, the lamps do not provide sufficient illumination in time, which can easily cause the driver's vision to be instantly blinded, thereby endangering personal safety. The current vehicle lighting system is not "intelligent" enough. It cannot think, analyze and learn with human thinking, nor can it actively adapt to match the driver's operating habits. Therefore, a vehicle auxiliary control system based on cloud computing resources and intelligent control is needed to integrate the vehicle body's lighting resources to save resources and improve driving safety.

[0004] The existing patent application with publication number CN111055759A discloses an intelligent oncoming light control system based on AI image recognition and navigation system, including: camera, navigation module, radar speedometer, intelligent vehicle terminal, cloud system, central control system, MCU, ECU; the camera, navigation module, and speed measurement module respectively collect image information of oncoming vehicles in the direction of vehicle travel, road condition information in front of the vehicle, vehicle travel speed and vehicle distance information, and upload the collected information to the cloud system through the intelligent vehicle terminal; the cloud system makes decisions based on the received information, generates control instructions and returns them to the intelligent vehicle terminal; the intelligent vehicle terminal sends the control instructions to the central control system, and the central control system converts the control instructions into instructions that can be recognized by the MCU, and controls the ECU to operate the light switching through the MCU. The light control system is insufficient in judging the driving conditions on complex roads, and has limited control integration for the light modules in the vehicle body, occupies a large amount of space and computing resources in the vehicle, and cannot interact with the driver to improve driving safety and comfort. Summary of the invention

[0005] One of the purposes of the present invention is to provide a vehicle lighting auxiliary control method based on ChatGPT, which solves the problem that the existing lighting is insufficient in judging complex driving environments.

[0006] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is as follows:

[0007] A ChatGPT-based automobile lighting auxiliary control method comprises the following steps:

[0008] S1. Acquiring external environment data: Image information of the driving environment is collected from the vehicle body. The image information is transmitted to the visual neural computing unit in the cloud through the visual neural unit on the vehicle body. The visual neural computing unit processes the image information to obtain a light operation instruction. The light operation instruction is transmitted to the headlight by the visual neural execution unit set on the vehicle body. The headlight is adjusted according to the light operation instruction.

[0009] S2. Obtaining control instructions: With the authorization of the driver, the vehicle body can receive the light control instructions issued by the driver, or automatically generate light control instructions according to the last driving record under the same circumstances. The light control instructions are transmitted to the headlights through the visual neural execution unit, and the headlights are adjusted according to the light control instructions;

[0010] S3. Comprehensive judgment: When abnormal image information appears in the driving environment collected by the vehicle body, the abnormal image information is transmitted to the visual neural computing unit through the visual neural unit. The visual neural computing unit obtains abnormal light operation instructions after processing the image information. The computing resource pool set up in the cloud obtains the abnormal light operation instructions, analyzes the abnormal light operation instructions and feeds back the results; if the cloud-based ChatGPT language model cannot distinguish the current abnormal instructions, it is transmitted to the driver in the form of voice, and the driver judges whether to execute the abnormal light operation instructions or input new light operation instructions to replace them. The environmental information and the driver's judgment can be combined to obtain safer and more effective light use instructions.

[0011] Furthermore, it also includes S4, diversification of lighting instructions: when abnormal image information appears in the driving environment collected by the vehicle body, the abnormal image information is transmitted to the visual neural computing unit through the visual neural computing unit, and the visual neural computing unit obtains abnormal lighting operation instructions after processing the image information. When the abnormal lighting operation instruction is similar to the previous abnormal image information, the computing resource pool obtains the abnormal lighting operation instruction, and transmits the abnormal lighting operation instruction to the driver in the form of voice. The driver determines whether the abnormal lighting operation instruction is executed according to the previous lighting operation instruction or inputs a new lighting operation instruction to replace it, which is used to iterate the lighting control instruction, so that the lighting control can change with the changes in the environment and the driver's operating habits, thereby improving driving safety and driving comfort.

[0012] Furthermore, the computing resource pool runs the ChatGPT language model for direct communication with people, and the lighting control is more in line with the driver's usage habits.

[0013] The second purpose of the present invention is to provide a car lighting auxiliary control system based on ChatGPT, which solves the problems of insufficient integration of existing lighting control systems and insufficient interaction with drivers.

[0014] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is as follows:

[0015] A car light auxiliary control system based on ChatGPT includes a computing resource pool and a visual neural computing unit arranged in the cloud, a visual neural exchange unit, a visual neural execution unit and a car light arranged on the car body, wherein the visual neural exchange unit and the visual neural execution unit collect image information of the car during driving and upload it to the visual neural computing unit, the visual neural computing unit processes the collected image information to obtain a light execution instruction, the light execution instruction is transmitted to the visual neural execution unit through the visual neural exchange unit, the visual neural execution unit controls the light emitting state of the car light, the computing resource pool is used to run an algorithm model, and the control of the light is divided into two parts, the car body and the cloud, thereby reducing the demand for car body hardware equipment.

[0016] Preferably, the visual neural exchange unit includes a neural network processor and a graphics processor for processing image data.

[0017] Preferably, the hardware architecture of the computing resource pool includes a central processing unit (CPU), a graphics processing unit (GPU), a tensor processing unit (TPU), a field programmable gate array (FPGA), an application-specific integrated circuit (ASIC), a data processing unit (DPU), a memory processing unit (MPU), a solid-state drive (SSD), a non-volatile memory express interface (NVMe), a dynamic random access memory (DDR), a network interface card (NIC), a pluggable factor replacement module (SFP), a cache (CacheMemory), a DMA controller and a neural network processing unit (NPU), which are used to support the learning and control needs of lighting.

[0018] Preferably, the vehicle lights include pixel headlights, matrix LED headlights, adaptive high beam (ADB), adaptive front lighting (AFS) and projection lights, which cover a wide range of vehicle lighting use and have a high degree of intelligence in the use of all vehicle lights.

[0019] Preferably, it also includes a wireless communication module arranged on the vehicle body, and the wireless communication module is one or more of a mobile network communication module, a WIFI module, a Zigbee module, a LoRa module, a NB-Iot module, a satellite communication module or a quantum communication module to improve the safety of the use of the lighting auxiliary control system.

[0020] Preferably, the learning methods of the ChatGPT language model include supervised learning, unsupervised learning and immersive learning, and the immersive learning types include convolutional neural networks, recurrent neural networks, long short-term memory networks, gated recurrent units, Transformer models, generative adversarial networks and autoencoders, which are used to improve learning efficiency and accuracy.

[0021] Furthermore, it also includes an input recognition module arranged on the vehicle body, which includes a voice recognition device, a face recognition device and a scene image sensor, and is used to obtain environmental information from multiple angles such as voice, face and scene image.

[0022] The beneficial effects of the present invention are:

[0023] (1) The ChatGPT-based automobile lighting auxiliary control method is equipped with a computing resource pool and a visual neural independent unit in the cloud for running large-scale model algorithms. According to the obtained environmental data, such as driving speed, road direction, lighting conditions, road conditions, etc., the method executes instructions for controlling the lighting state. At the same time, it can also receive lighting operation instructions from the driver and transmit the instructions for controlling the lights to the visual neural exchange unit on the vehicle body. The visual neural execution unit controls the lighting state of the lights to realize lighting on complex roads. By arranging a computing resource pool in the cloud, the hardware requirements for the vehicle body can be reduced, and different types of vehicle body lights can be integrated to reduce the occupation of vehicle body space and computing resources, thereby improving the intelligence of the vehicle body.

[0024] (2) The ChatGPT-based automobile lighting auxiliary control system has a ChatGPT language model in its computing resource pool. The driver can input voice control signals in the car to control the lighting status of the entire body. The ChatGPT language model analyzes and collects road conditions and upcoming traffic conditions, and can communicate with the driver in advance. It can also collect driving environment data and learn lighting control from the driver to iterate lighting control, ensure that lighting use is adapted to the current situation, and improve driving safety and comfort.

[0025] (3) The ChatGPT-based automobile lighting auxiliary control method interacts with the driver through voice or text via the ChatGPT language model. It can actively communicate with the driver when the algorithm judgment in the computing resource pool is abnormal, and comprehensively control the use status of the lights based on environmental conditions and the driver's judgment, thereby improving driving safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a functional composition diagram of the automobile lighting auxiliary control system based on ChatGPT provided by the present invention;

[0027] Figure 2 The operation logic diagram of the automobile lighting auxiliary control system based on ChatGPT provided by the present invention;

[0028] Figure 3 The hardware architecture diagram of the computing resource pool of the ChatGPT-based automobile lighting auxiliary control system provided by the present invention;

[0029] Figure 4 The operation flow chart of the automobile lighting auxiliary control system based on ChatGPT provided by the present invention;

[0030] Figure 5 A diagram showing a learning method of the ChatGPT language model of the ChatGPT-based automobile light auxiliary control system provided by the present invention;

[0031] Figure 6 This is an information transmission flow chart of the automobile lighting auxiliary control system based on ChatGPT provided by the present invention. DETAILED DESCRIPTION

[0032] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0033] Embodiment 1

[0034] like Figure 1-6 As shown, this embodiment discloses a vehicle lighting auxiliary control method based on ChatGPT, comprising the following steps:

[0035] S1. Acquiring external environment data: The image information of the driving environment is collected from the vehicle body. The image information is transmitted to the visual neural computing unit in the cloud through the visual neural unit on the vehicle body. The visual neural computing unit processes the image information to obtain the light operation instruction. The light operation instruction is transmitted to the headlight by the visual neural execution unit set on the vehicle body. The headlight is adjusted according to the light operation instruction. The driving environment information collected by the vehicle body controls the light action, which has a high degree of intelligence and reduces the operating burden of the driver.

[0036] S2. Obtaining control instructions: With the driver's authorization, the vehicle body can receive the lighting control instructions issued by the driver, or automatically generate lighting control instructions based on the last driving record under the same circumstances. The lighting control instructions are transmitted to the headlights through the visual neural execution unit, and the headlights are adjusted according to the lighting control instructions. The lights can be directly controlled by the driver or autonomously controlled by the ChatGPT language model without the driver's intervention, providing a wider range of control requirements, and each time the lighting control instructions are executed, the lighting control instructions in that situation can be memorized for quick call when encountering the same situation;

[0037] S3. Comprehensive judgment: When abnormal image information appears in the driving environment collected by the vehicle body, the abnormal image information is transmitted to the visual neural computing unit through the visual neural unit. The visual neural computing unit obtains abnormal light operation instructions after processing the image information. The computing resource pool set up in the cloud obtains the abnormal light operation instructions, analyzes the abnormal light operation instructions and feeds back the results; if the cloud-based ChatGPT language model cannot distinguish the current abnormal instructions, it is transmitted to the driver in the form of voice, and the driver determines whether to execute the abnormal light operation instructions or input new light operation instructions to replace them, so that the lights can be used safely in complex driving environments.

[0038] Furthermore, it also includes S4, diversification of lighting instructions: when abnormal image information appears in the driving environment collected by the vehicle body, the abnormal image information is transmitted to the visual neural computing unit through the visual neural computing unit, and the visual neural computing unit obtains abnormal lighting operation instructions after processing the image information. When the abnormal lighting operation instruction is similar to the previous abnormal image information, the computing resource pool obtains the abnormal lighting operation instruction, and transmits the abnormal lighting operation instruction to the driver in the form of voice. The driver determines whether the abnormal lighting operation instruction is executed according to the previous lighting operation instruction or inputs a new lighting operation instruction to replace it. Through iteration, the lighting can meet the driving needs in any environment and can change with the external environment and the driver's operating habits.

[0039] See also Figure 3 Furthermore, the computing resource pool runs the ChatGPT language model, which can directly communicate with the driver through voice or text to meet the driver's lighting needs. At the same time, it can analyze and calculate the environmental information data obtained on the visual neural computing unit, and the visual neural execution unit drives the headlights to execute lighting instructions; when the algorithm model in the cloud computing resource pool cannot handle the road conditions or an abnormal situation occurs, the ChatGPT language module can communicate with the driver to handle the abnormality and clarify the lighting execution instructions.

[0040] Embodiment 2

[0041] The present embodiment also discloses a vehicle light auxiliary control system based on ChatGPT, which can be used to implement a vehicle light auxiliary control method, including a computing power resource pool and a visual neural computing power unit arranged in the cloud, a visual neural exchange unit, a visual neural execution unit and vehicle lights arranged on the vehicle body, the visual neural exchange unit and the visual neural execution unit collect image information of the vehicle during driving and upload it to the visual neural computing power unit, the visual neural computing power unit processes the collected image information to obtain a light execution instruction, the light execution instruction is transmitted to the visual neural execution unit through the visual neural exchange unit, the visual neural execution unit controls the light-emitting state of the vehicle lights, the computing power resource pool is used to run the algorithm model, the vehicle body is used to collect road environment information and feed it back to the cloud, the cloud calculates the light execution instruction and transmits it to the vehicle body to control the light-emitting state of the vehicle lights, different vehicle lights on the vehicle body can be integrated to reduce the demand for computing power, thereby saving vehicle body space and hardware costs, and the integrated vehicle lights can meet the driving light requirements outside the vehicle as well as the lighting and indication requirements inside the vehicle.

[0042] Among them, the computing resource pool is the core brain of cloud computing. All algorithm learning of the headlight lighting system is carried out in the cloud. Before the headlights perform lighting actions, the visual neural exchange unit collects environmental data from the vehicle body and transmits it to the computing resource pool in real time without the need for computational processing. By reducing the intermediate processes, the efficiency of information processing is improved.

[0043] Furthermore, the visual neural exchange unit includes a neural network processor and a graphics processor for processing image data for analysis, induction and comparison, decomposing the image into recognizable data and making judgments after comparing the data with internal data.

[0044] Furthermore, the hardware architecture of the computing resource pool includes a central processing unit (CPU), a graphics processing unit (GPU), a tensor processing unit (TPU), a field programmable gate array (FPGA), an application-specific integrated circuit (ASIC), a data processing unit (DPU), a memory processing unit (MPU), a solid-state drive (SSD), a non-volatile memory express interface (NVMe), a dynamic random access memory (DDR), a network interface card (NIC), a pluggable factor replacement module (SFP), a cache memory, a direct memory access DMA controller, and a neural network processing unit (NPU).

[0045] Among them, the field programmable gate array (FPGA) adopts a parallel architecture and data stream processing to realize algorithm acceleration or be used as a hardware acceleration card to improve data processing efficiency; the application-specific integrated circuit (ASIC) has no redundant functions compared to the general-purpose processing circuit, but has the characteristics of high computing power and low latency, and can be used for cryptocurrency, data processing, network processing, immersive learning and inference learning, and can be used here to improve learning efficiency; the data processing unit (DPU) accelerates data processing and optimizes the network and storage; the memory processing unit (MPU) performs calculations and decisions, including executing instructions and processing data, and can be used as a scheduling system for operating systems and tasks. It has strong versatility in computing power and a rich instruction set; the solid-state drive (SSD) provides fast data reading and writing capabilities, stores model training data and program code, can improve I / O bottlenecks, and improve system System startup acceleration, speed up application loading, improve server performance, and support big data processing; Non-volatile Memory Express (NVMe), a high-performance storage interface, is used to connect solid-state drives (SSDs) to provide fast data transmission; Dynamic Random Access Memory (DDR), provides high-speed data transmission, improves bandwidth, reduces CPU waiting time, and responds quickly to cache misses, with multi-tasking processing capabilities; Network Interface Card (NIC), provides network connection, supports high-speed data exchange, reduces the computing burden of the central processing unit (CPU), supports virtual local area network (VLAN), and can perform network monitoring and redundancy management; Small Pluggable Factor Converter (SFP), supports optical fiber and Ethernet communications, is used for high-speed network connection, has the characteristics of pluggable and flexible network, and can also perform network monitoring and performance optimization; Cache (Cache The main memory provides fast-access temporary storage, reduces the latency of the processing unit accessing the main memory, and increases the data processing speed by prefetching data and prefetching instructions. The DMA controller is used for fast startup and acceleration, improves application performance, reduces interruptions to the central processing unit (CPU), and performs high-speed network transmission and fast storage access. The cloud-based neural network processing unit (NPU) is used for neural network computing optimization, accelerated training and reasoning, and has the characteristics of high throughput and low latency, assisting machine learning and autonomous driving learning.

[0046] The workflow of the car light is as follows:

[0047] First, the neural network processing unit (NPU) or the driver proposes a task requirement and sends it to the central processing unit (CPU), which breaks down the task requirement and converts it into specific execution instructions. The central processing unit (CPU) then distributes the execution instructions to the domain controller (BDCU) of the vehicle body. After receiving the execution instructions, the domain controller drives the underlying driver to perform the action; the underlying driver includes radar, camera, infrared sensor, camera and other actuators that can collect external environmental information, which are used to collect driving information such as images, light intensity, road conditions, traffic conditions, and driving speed; after receiving the image information and environmental information obtained by the domain controller (BDCU), the central processing unit (CPU) uploads it to the On the cloud-based neural network processing unit (NPU), scene recognition, image analysis, and speech analysis are performed by the neural network processing unit (NPU). The cloud-based neural network processing unit (NPU) will distribute the recognized image data to the graphics processing unit (GPU) and tensor processing unit (TPU) for calculation in the form of task requirements. After the graphics processing unit (GPU) and tensor processing unit (TPU) complete the image analysis calculation, they will uniformly pass the calculated results to the central processing unit (CPU), and then send them to the domain controller (BDCU) via the central processing unit (CPU). The domain controller BDCU executes the control instructions of the headlights through the underlying driver to control the working status of the headlights. In the actual operation of the headlights, cache memory and DMA controller are used to transfer data to reduce data latency of the server.

[0048] Preferably, the vehicle lights include pixel headlights, matrix LED headlights, adaptive high beam (ADB), adaptive front lighting (AFS) and projection lights, which can control the working status of all lights in the vehicle.

[0049] Preferably, it also includes a wireless communication module arranged on the vehicle body, and the wireless communication module is one or more of a mobile network communication module, a WIFI module, a Zigbee module, a LoRa module, a NB-Iot module, a satellite communication module or a quantum communication module, so that the vehicle body can still work normally when there is no signal or the signal strength is low, and the wide coverage of satellite communication and the stability of quantum communication are utilized to ensure the security and response speed of data transmission.

[0050] Preferably, it also includes an input recognition module arranged on the vehicle body, which includes a voice recognition device, a face recognition device and a scene image sensor, which is used to input environmental information inside and outside the vehicle, enhance the ability to perceive the external environment, and thus improve driving comfort and safety.

[0051] Preferably, the learning methods of the ChatGPT language model include supervised learning, unsupervised learning and immersive learning. The immersive learning types include convolutional neural networks, recurrent neural networks, long short-term memory networks, gated recurrent units, Transformer models, generative adversarial networks and autoencoders. Through a variety of learning methods, the distinction of the external environment can be improved, the ability to cope with complex driving environments can be improved, and driving safety can be improved by controlling lights.

[0052] Among them, the ChatGPT language model is divided into a learning module and an execution module. The learning module can accurately identify the working status of the light, distinguish the light intensity, detect the driving conditions and understand the driver's lighting instructions; the execution module can more directly respond and manipulate according to the road conditions, and can execute the corresponding light language according to the driver's voice commands, such as turning, reversing, and overtaking light prompts. The input recognition module of the car body recognizes voice, face and scene, so that the ChatGPT language model can more conveniently observe the lighting needs inside and outside the car.

[0053] The specific operation process of the ChatGPT language model is as follows:

[0054] Supervision learning method: training how to judge whether the lights are working properly

[0055] Data: Collect a large amount of image data of the working status of car lights in different conditions, and annotate these images, such as labels such as "normal lighting", "insufficient brightness" and "damaged and not lit".

[0056] Task: By learning these clearly labeled image data, gradually grasp the characteristics of car lights in different states.

[0057] Application: After training, the data model can accurately determine whether the lighting status of a car light belongs to "normal lighting", "insufficient brightness" or "damaged and not lit".

[0058] Supervision learning method: training how to optimize the lighting control of vehicle lights.

[0059] Data: Collect the optimal parameter settings for vehicle lighting under different road conditions (such as urban roads, highways, rural roads, etc.) and different weather conditions (sunny, rainy, foggy, etc.) as well as the corresponding scene labels.

[0060] Task: Gradually grasp the road conditions of different states by learning these image data with clear labels.

[0061] Application: After training, the algorithm model can determine the current road and weather conditions, and automatically adjust the lighting mode and brightness of the headlights to achieve the best lighting effect and improve driving safety.

[0062] Unsupervised Learning: Scene Recognition

[0063] Data: Label data is not pre-annotated on the image data. For example, a large amount of image data of car lighting in different scenes is collected but no labeling is performed on the data.

[0064] Task: The lighting auxiliary control system can automatically cluster similar lighting image data by analyzing the characteristics of the input image data. For example, it can cluster the lighting images of vehicles with similar brightness, similar illumination angles, or in specific scenes (such as urban roads at night and mountain curves) into one category.

[0065] Application: After successful learning, different types of headlight lighting patterns can be discovered, and when encountering new lighting conditions in the future, they can be classified into corresponding categories according to their characteristics, thus providing a reference for optimizing headlight lighting. In addition, some potential patterns can be discovered by analyzing the changes in headlight lighting data over a long period of time. For example, it may be found that the brightness of the headlights gradually decreases with the increase in usage time, or the change patterns of headlight performance in different seasons and temperatures. This can predict possible problems with the headlights in advance so that maintenance and adjustments can be made in a timely manner.

[0066] Immersive Learning:

[0067] Convolutional Neural Networks (CNNs)

[0068] Scenario 1: When using adaptive high beam (ADB), convolutional neural networks can be used to identify vehicles and pedestrians ahead. Images of the road ahead are collected by a camera installed on the vehicle, and the images are input into a convolutional neural network for processing. The network can automatically extract features in the image, such as the shape and edges of objects, to determine whether there are vehicles or pedestrians ahead. When a vehicle or pedestrian is detected ahead, the adaptive high beam (ADB) can automatically adjust the range and intensity of illumination to avoid glare to vehicles and pedestrians ahead.

[0069] Scenario 2: When using pixel headlights, convolutional neural networks can be used to achieve dynamic lighting effects. For example, the lighting pattern of pixel headlights can be automatically adjusted according to the road conditions ahead, such as adding side lighting on curves and adding high beam lighting on highways. By training on a large number of images of different road scenes, convolutional neural networks can learn the best lighting patterns for different scenes.

[0070] Recurrent Neural Networks (RNNs)

[0071] Scenario 3: When using adaptive front lighting (AFS), the recurrent neural network can be used to predict the vehicle's driving trajectory. By analyzing the vehicle's speed, steering angle and other sensor data, the recurrent neural network can predict the vehicle's driving direction and position in the future. Based on the prediction results, the adaptive front lighting (AFS) can automatically adjust the direction of the light to provide good lighting for the vehicle in advance.

[0072] Scenario 4: In projection headlights, recurrent neural networks can be used to achieve dynamic projection effects. For example, the content and brightness of the projection pattern can be automatically adjusted according to the vehicle's speed and changes in the surrounding environment. For example, speed limit signs and pedestrian crossing reminders can be projected on urban roads, and lane lines and road conditions ahead can be projected on highways.

[0073] Long Short-Term Memory Networks (LSTMs)

[0074] Scenario 5: When using digital projectors (DLP), long short-term memory networks can be used to memorize and learn different lighting scenarios. For example, when a vehicle is driving under different road conditions and weather conditions, the digital projector (DLP) can automatically adjust the lighting mode. LSTMs can remember the lighting settings of similar scenarios encountered before and make appropriate adjustments based on the current situation. This can improve the adaptability and intelligence of the headlights.

[0075] Scenario 6: In artificial dialogue systems, LSTMs can be used to understand user instructions and questions. When a user is talking to a car lighting system, LSTMs can analyze the user's language sequence, understand the user's intention, and give corresponding answers and operation suggestions. For example, a user can use voice commands to ask the headlights to turn on a specific lighting mode, and LSTMs can recognize the user's instructions and control the headlights to make corresponding adjustments.

[0076] Gated Recurrent Units (GRUs)

[0077] Scenario 6: When using adaptive high beam (ADB), the gated recurrent unit can be used to monitor the motion state of the vehicle in front in real time. By analyzing the image sequence collected by the camera, GRUs can predict the changes in the speed and driving direction of the vehicle in front. Based on these prediction results, the adaptive high beam (ADB) can adjust the lighting range and intensity in time to ensure that it does not cause glare to the vehicle in front while providing good lighting for its own driving.

[0078] Scenario 7: In terms of recognition, GRUs can be used to identify different traffic signs and signals. By analyzing the images collected by the camera, GRUs can identify the shape, color and meaning of traffic signs, as well as the status of traffic lights. This information can be used to automatically adjust the lighting mode of the headlights, such as automatically reducing the brightness of the headlights when encountering a red light, and increasing side lighting when passing through an intersection.

[0079] Transformer Model

[0080] Scenario 8: When using pixel headlights and digital projectors (DLP), the Transformer model can be used to achieve more complex lighting effects. For example, by training a large number of images of different scenes, the Transformer model can learn how to generate the best lighting pattern according to different scene requirements. The Transformer model can consider multiple factors at the same time, such as road shape, vehicle position, surrounding environment, etc., to generate more accurate and personalized lighting solutions.

[0081] Scenario 9: In artificial dialogue systems, the Transformer model can be used to improve language understanding and generation capabilities. When a user is talking to a car lighting system, the Transformer model can better understand the user's questions and instructions and generate more accurate and natural responses. At the same time, the Transformer model can also continuously optimize its responses based on user feedback to improve user experience.

[0082] Generative Adversarial Networks (GANs)

[0083] Scenario 10: In vehicle lighting design, generative adversarial networks can be used to generate new lighting effects and patterns. Designers can input some basic design requirements and parameters to let GANs generate a variety of different lighting solutions. Designers can then select and optimize based on these generated solutions, thereby improving design efficiency and innovation capabilities.

[0084] Scenario 11: In data enhancement, GANs can be used to generate more training data. For example, for the training of pixel headlights and digital projectors (DLP), a large amount of image data of different scenes is required. By using GANs to generate some synthetic image data, the diversity of training data can be increased and the generalization ability of the model can be improved.

[0085] Autoencoder Model

[0086] Scenario 12: In the fault detection of the headlight system, the autoencoder can be used to detect abnormal data. By training the headlight data under normal working conditions, the autoencoder can learn the characteristic distribution of normal data. When a fault occurs, the input data will be different from the characteristic distribution of normal data, and the reconstruction error of the autoencoder will increase. By monitoring the reconstruction error, the fault of the headlight system can be discovered in time, and corresponding repairs and maintenance can be performed.

[0087] Scenario 13: In terms of data processing, the autoencoder can be used to compress data in the vehicle lighting system. For example, the control signals of pixel headlights and digital projectors (DLP) can be compressed using the autoencoder to reduce the cost of data transmission and storage. At the same time, when the data is needed, the decoder can be used to restore the compressed data to the original signal to ensure the normal operation of the lighting system.

[0088] More preferably, it also includes a domain controller and an underlying driver arranged on the vehicle body, and the underlying driver includes radar, camera, infrared sensor and camera, which are used to collect the speed of the car and the traffic conditions of vehicles on the same road section to eliminate potential risks.

[0089] See also Figure 5 and Figure 6The automobile light auxiliary control system in the present invention relies on the ChatGPT language model for implementation, which is different from the cloud computing and AI image recognition road conditions in the prior art. The ChatGPT language model is used to control the lights with human thinking habits, and mapping is performed by simulating the visual neural units of the human body. In the process of learning light control, the car focuses on learning and mastering the lighting control of what brightness to provide in different situations, such as the action and intensity characteristics of the current car lights. When encountering the same situation next time, it is determined whether to take the same control instructions or find better light control instructions, that is, a problem can be solved in multiple different ways. The core of the lighting assistance control system is the ChatGPT language model, which is divided into a learning module and an execution module. The learning module is a process from scratch, from recognition to familiarity, and from existing to innovation; while the execution module is a process from slow to fast, from one method of control to multiple methods of control; whether in learning or execution mode, the ChatGPT language model will send the lighting execution command to the body input recognition module based on cloud and local data conversion. The body input recognition module uses three modules: voice recognition, face recognition, and scene recognition. The voice recognition module allows the ChatGPT language model to communicate directly with the driver, and through the dialogue, it can better understand the driver's lighting needs and improve driving comfort; the face recognition module allows the ChatGPT language model to communicate directly with the driver, and through the face recognition module, it can The atGPT language model can better observe whether the driver is in a state of fatigue or bad mood. At this time, active measures such as turning on the driver's preferred ambient light or playing music can be taken to relieve the driver's fatigue and soothe his emotions, so as to improve driving safety. Through the scene recognition module, the car lighting auxiliary control system can better observe the road conditions outside the car, such as external ambient temperature, light brightness, climate humidity, distinguishing between moving objects and fixed objects, road surface flatness, color distinction, and vehicle speed judgment. After the body input recognition module is recognized, the result data will be uploaded to the ChatGPT language model, and after calculation, the corresponding light control instructions will be sent to the body input recognition module, and the body input recognition module will then feed back to the headlight control module to control the headlight lighting. Figure 6 Taking digital projection lights (DLP) as an example, during actual driving, the ChatGPT language model decodes and calculates the voice, facial features, and internal / external environment collected by the vehicle input recognition module through cloud and local data conversion, and feeds it back to the vehicle input recognition module. After passing through the headlight control module, the headlight lighting is controlled. If a pedestrian is found crossing the road in front, the ChatGPT language model will determine that there may be a dangerous collision ahead based on previous learning experience. In order to avoid this danger, the vehicle's BDCU domain controller will slow down and control the safe braking distance, and then transmit hazard lights on the ground to remind pedestrians that there are oncoming vehicles passing by and to pay attention to avoid them.

[0090] The present invention can effectively save the hardware computing resources and software computing resources of the product by centrally utilizing the computing resources of the cloud instead of loading the computing resources onto a single car. The visual neural network unit is introduced so that the lighting of the car lights can be adjusted in real time according to the on-site environmental conditions, just like humans. In addition, by shortening the control process of the car lights, all the car lights share the same cloud server and are uniformly managed by the ChatGPT language model and the visual neural computing system, which can reduce costs and increase efficiency, shorten the R&D cycle, enhance product compatibility, and make it easier to expand capacity on the existing basis. The ChatGPT language model can understand human language, communicate with the driver and learn the use of lights, and has the ability of self-deduction and collection and classification, which greatly reduces the driver's attention occupation and makes driving more comfortable and safe.

[0091] According to the disclosure and teaching of the above description, those skilled in the art to which the present invention belongs may also change and modify the above embodiments. Therefore, the present invention is not limited to the specific embodiments disclosed and described above, and some modifications and changes of the present invention should also fall within the protection scope of the claims of the present invention. In addition, although some specific terms are used in this specification, these terms are only for the convenience of description and do not constitute any limitation to the present invention.

Claims

1. A vehicle lighting auxiliary control method based on ChatGPT, characterized in that: The following steps are involved: S1. Acquiring external environment data: Image information of the driving environment is collected from the vehicle body. The image information is transmitted to the visual neural computing unit in the cloud through the visual neural unit on the vehicle body. The visual neural computing unit processes the image information to obtain a light operation instruction. The light operation instruction is transmitted to the headlight by the visual neural execution unit set on the vehicle body. The headlight is adjusted according to the light operation instruction. S2. Obtaining control instructions: With the authorization of the driver, the vehicle body can receive the light control instructions issued by the driver, or automatically generate light control instructions according to the last driving record under the same circumstances. The light control instructions are transmitted to the headlights through the visual neural execution unit, and the headlights are adjusted according to the light control instructions; S3. Comprehensive judgment: When abnormal image information appears in the driving environment collected by the vehicle body, the abnormal image information is transmitted to the visual neural computing unit through the visual neural unit. The visual neural computing unit obtains abnormal light operation instructions after processing the image information. The computing resource pool set up in the cloud obtains the abnormal light operation instructions, analyzes the abnormal light operation instructions and feeds back the results; if the cloud-based ChatGPT language model cannot distinguish the current abnormal instructions, it is transmitted to the driver in the form of voice, and the driver determines whether to execute the abnormal light operation instructions or input new light operation instructions to replace them.

2. The ChatGPT-based automobile lighting auxiliary control method according to claim 1 is characterized in that: It also includes step S4, diversification of lighting instructions: when abnormal image information appears in the driving environment collected by the vehicle body, the abnormal image information is transmitted to the visual neural computing unit through the visual neural computing unit, and the visual neural computing unit obtains abnormal lighting operation instructions after processing the image information. When the abnormal lighting operation instruction is similar to the previous abnormal image information, the computing resource pool obtains the abnormal lighting operation instruction, and transmits the abnormal lighting operation instruction to the driver in the form of voice, and the driver determines whether the abnormal lighting operation instruction is executed according to the previous lighting operation instruction or inputs a new lighting operation instruction to replace it.

3. The ChatGPT-based automobile lighting auxiliary control method according to claim 1 or 2, characterized in that: The ChatGPT language model is run in the computing power resource pool.

4. An automobile lighting auxiliary control system based on ChatGPT, characterized in that: It includes a computing resource pool and a visual neural computing unit arranged in the cloud, a visual neural exchange unit, a visual neural execution unit and a headlight arranged on the vehicle body. The visual neural exchange unit and the visual neural execution unit collect image information of the vehicle while it is driving and upload it to the visual neural computing unit. The visual neural computing unit processes the collected image information to obtain a light execution instruction. The light execution instruction is transmitted to the visual neural execution unit through the visual neural exchange unit. The visual neural execution unit controls the light-emitting state of the headlight. The computing resource pool is used to run the algorithm model.

5. The ChatGPT-based automobile lighting auxiliary control system according to claim 4 is characterized in that: The visual neural exchange unit includes a neural network processor and a graphics processor.

6. The ChatGPT-based automobile lighting auxiliary control system according to claim 4 is characterized in that: The hardware architecture of the computing power resource pool includes a central processing unit (CPU), a graphics processing unit (GPU), a tensor processing unit (TPU), a field programmable gate array (FPGA), an application-specific integrated circuit (ASIC), a data processing unit (DPU), a memory processing unit (MPU), a solid-state drive (SSD), a non-volatile memory express interface (NVMe), a dynamic random access memory (DDR), a network interface card (NIC), a pluggable factor replacement module (SFP), a cache (Cache Memory), a DMA controller and a neural network processing unit (NPU).

7. The ChatGPT-based automobile lighting auxiliary control system according to claim 4 is characterized in that: The vehicle lights include pixel headlights, matrix LED headlights, adaptive high beam (ADB), adaptive front lighting (AFS), and projection lights.

8. The ChatGPT-based automobile lighting auxiliary control system according to any one of claims 4 to 7, characterized in that: It also includes a wireless communication module arranged on the vehicle body, and the wireless communication module is one or more of a mobile network communication module, a WIFI module, a Zigbee module, a LoRa module, a NB-Iot module, a satellite communication module or a quantum communication module.

9. The ChatGPT-based automobile lighting auxiliary control system according to claim 8, characterized in that: The learning methods of the ChatGPT language model include supervised learning, unsupervised learning and immersive learning, and the immersive learning types include convolutional neural networks, recurrent neural networks, long short-term memory networks, gated recurrent units, Transformer models, generative adversarial networks and autoencoders.

10. The ChatGPT-based automobile lighting auxiliary control system according to claim 9, characterized in that: It also includes an input recognition module arranged on the vehicle body, and the input recognition module includes a voice recognition device, a face recognition device and a scene image sensor.

Images