Annular information acquisition terminal equipment for self-driving automobile
By combining lidar, 360 camera and high-precision camera in the ring information acquisition terminal equipment of autonomous driving vehicles, combined with deep learning and virtual three-dimensional models, the problem of insufficient accuracy of ring information acquisition data is solved, and the accuracy of target recognition and avoidance is improved , enhance driving safety and personnel comfort, and optimize the rollover neural network through the cloud, improving the output effect of driving instructions.
Patent Information
- Application Number
- CN202410970357.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-18
- Publication Date
- 2024-11-26
AI Technical Summary
During driving, due to insufficient accuracy of ring information collection data, the target recognition speed is slow and the problem of untimely avoidance.
A ring information acquisition terminal device for autonomous driving vehicles is designed, including data acquisition module, data processing module, data control module, data matching module, deep learning module and cloud. Environmental data is collected through lidar, 360 camera and high-precision camera. Combining deep learning and virtual three-dimensional models, optimize sensor acquisition methods and data processing to generate accurate driving instructions.
It improves the accuracy of target recognition and the accuracy of avoidance during automatic driving, enhances driving safety and personnel comfort, and optimizes the rollover neural network in the cloud to improve the output effect of driving commands.
Smart Images

Figure CN119030999A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of autonomous driving technology, and in particular to a ring-shaped information collection terminal device for an autonomous driving vehicle. Background Art
[0002] The ring information collection terminal device of the autonomous driving vehicle is a device used to collect surrounding environment information to support the decision-making of the autonomous driving vehicle. The device usually includes multiple sensors, such as cameras, radars, lidars, and ultrasonic sensors, which are used to identify and track surrounding roads, vehicles, pedestrians and other objects in real time;
[0003] However, the accuracy of the collected data of the car's circular information has a great impact on the car's driving process, which may cause the car to recognize targets on the road slowly, resulting in the car's slow avoidance when driving due to slow target recognition. For this reason, we propose a terminal device for collecting circular information of autonomous driving vehicles. Summary of the invention
[0004] The purpose of the present invention is to provide a ring-shaped information collection terminal device for an autonomous driving vehicle.
[0005] In order to solve the problems raised in the above background technology, the present invention provides the following technical solutions: a ring-shaped information collection terminal device for an autonomous driving vehicle, wherein the collection terminal device communicates with the vehicle driving system, and the collection terminal device includes a data collection module, a data processing module, a data control module, a data matching module, a deep learning module, a data transmission module and a cloud;
[0006] The data acquisition module includes an environment acquisition unit and a data acquisition unit. The data acquisition unit acquires environment data around the autonomous driving vehicle through sensors. The environment data acquired by the data acquisition unit includes radar data and shooting data, and transmits the data to the data processing module. The environment acquisition unit is used to acquire environment data around the data acquisition unit. The environment data acquired by the environment acquisition unit includes light data and weather data, and transmits the data to the data processing module.
[0007] The data processing module includes a processing unit and an abnormal unit. After receiving the data from the data acquisition unit, the processing unit identifies the scanned target according to the radar data, and returns the target data of the inaccurate scanning position to the data control module. After receiving the environmental data transmitted by the environmental acquisition unit, the abnormal unit processes the light data and weather data collected by the environmental acquisition unit, and then generates a corresponding control instruction according to the processing result, and then transmits the control instruction to the data control module.
[0008] The data control module is used to adjust the start and stop of the sensor in the data acquisition unit, thereby adjusting the acquisition mode of the data acquisition unit, and converting the acquisition mode into instruction data to be transmitted to the data acquisition module;
[0009] The data matching module is used to match the identified target data with the data in the acquisition terminal device;
[0010] The deep learning module is used to determine the safety level between the vehicle motion state and the target data;
[0011] The data transmission module is used to transmit the driving instruction data to the automobile driving system;
[0012] The cloud is used to receive data around the car while driving;
[0013] The data acquisition unit includes a laser radar, a 360 camera and a high-precision camera, and uses the laser radar and the 360 camera to identify roads and obstacles around the car, and transmits the identified data to the data processing module. The processing unit is used to identify the collected road condition data, and transmits the target position data that is not accurately identified to the data control module. The data control module controls the high-precision camera to shoot the unidentified target, and transmits the photographed target data to the processing unit. The processing unit is used to match the photographed target with the inaccurately identified target, and identify the target again.
[0014] As a further solution of the present invention: the data matching module includes a data storage unit, a storage matching unit and a judgment unit, the data storage unit is used to store virtual model data associated with road conditions, the storage matching unit matches the virtual model data in the data storage unit according to the target data identified by the data processing module, and transmits the successfully matched data to the judgment unit;
[0015] The deep learning module includes an optimization instruction unit and a convolutional neural network. The convolutional neural network is used to determine the safety level of the current car in the automatic driving situation, and determine the driving state of the car according to the safety level, and generate corresponding driving instructions, and transmit the driving instructions to the optimization instruction unit, the data transmission module and the cloud;
[0016] After receiving the environmental data, collected data and data from each module, the cloud generates and stores corresponding logs.
[0017] As a further solution of the present invention: the judgment unit simulates the driving data of the car when driving according to the matching virtual model, and judges the current driving state of the car according to the current driving data of the car, and transmits the judgment data to the deep learning module; the optimization instruction unit is used to optimize the driving state of the car, and obtain the best optimization method, and transmit the best optimization method data to the convolutional neural network; the environment acquisition unit includes a light sensor and a meteorological sensor, and the light sensor is used to obtain the light intensity when the 360 camera and the high-precision camera are shooting, and the light intensity data is transmitted to the abnormal unit, and the abnormal unit adjusts the angle of the camera according to the light intensity and fills the image with light to ensure the accuracy of the 360 camera and the high-precision camera in shooting the target.
[0018] As a further solution of the present invention: the processing unit combines radar data, 360 camera and high-precision camera, adjusts the high-precision camera to lock the target through the data processing module, and uses the 360 camera and high-precision camera to collect data of different angles of the target. The abnormal unit selects different data collection methods according to different weather conditions, and transmits the collection method data to the data control module. The data control module adjusts the number of sensors used according to the data collection method. When foggy weather and heavy rain affect the road view, the data control module transmits adjustment instructions to the data collection module, and enables laser radar, infrared camera, thermal camera and millimeter wave radar to collect road condition information in real time, ensuring that the car can avoid obstacles during driving and avoiding collisions when the field of vision is blocked.
[0019] As a further solution of the present invention: after the data processing module recognizes the target data, the data matching module matches the target data with the target virtual three-dimensional model in the data storage unit, and assembles the matched road condition model and target model into a virtual three-dimensional model. The judgment unit judges the danger level of the car driving according to the virtual three-dimensional model, and the danger level is divided into three levels: low, medium and high. When the judgment unit is a low danger level, the optimization instruction unit intervenes, and after the convolutional neural network generates an automatic driving instruction, the driving instruction is optimized. When the judgment unit is a medium danger level or a high danger level, the optimization instruction unit does not intervene in the automatic driving instruction of the convolutional neural network. The driving instruction generated by the convolutional neural network is transmitted to the car driving system by the data transmission module, and then the optimization instruction unit generates optimization data according to the driving status, and transmits the optimization data to the convolutional neural network.
[0020] As a further solution of the present invention: the optimization instruction unit determines the reward value that needs to be adjusted under the low risk level state through a set reward formula, and the specific reward formula is as follows:
[0021] Xzg =f1·Z i +f2·Q i +f3·W i , (f1+f2+f3=1)
[0022] Among them, X zg represents the reward value for optimizing driving instructions, Z i Indicates the braking force, Q i Indicates the driving stability value of the car, W i represents the comfort value of the person in the car, f1, f2 and f3 represent Z i , Q i and W i The weight coefficient of the optimization instruction unit is that when the danger level is at the medium danger level and the high danger level, the reward value that the convolutional neural network needs to update is obtained through the reward formula. The specific reward formula is:
[0023] X zf =f5·Z i +f6·Q i +f7·W u +f8·E u , (f5+f6+f7+f8=1)
[0024] Among them, X zf Represents the reward value of updating the convolutional neural network, W u Indicates the car loss ratio, E u represents the personnel safety value, f5, f6, f7 and f8 represent Z i , Q i , W u and E u The weight coefficient is used to improve the output of the convolutional neural network using the reward formula, thereby optimizing the output of the convolutional neural network for medium and high risk levels.
[0025] As a further solution of the present invention: a simulation calculation unit is set in the deep learning module, and the simulation calculation unit is used to obtain a virtual three-dimensional model and simulate the driving state of the car under the physical state, so as to ensure the feasibility of the current optimization instruction and transmit the feasible driving instruction data to the convolutional neural network.
[0026] As a further solution of the present invention: a hot patch is set in the convolutional neural network. After receiving the data of the self-driving car in the cloud, the data with a reward value higher than 95% in the optimization plan is transmitted to the main server. The main server generates hot patch data according to the optimization plan and updates the convolutional neural network in the collection terminal device in the networked state.
[0027] As a further solution of the present invention: after receiving the driving log of the car, the cloud generates a visual list of the driving log, which is convenient for personnel to connect to the cloud through the APP to retrieve the car driving data.
[0028] By adopting the above technical solution, compared with the prior art, the beneficial effects of the present invention are:
[0029] 1. The present invention can ensure the accuracy of the data acquisition unit in collecting the environmental data around the car through the setting of the environmental acquisition unit, use the environmental data to correct the data collected by the sensor in the data acquisition unit to ensure the data acquisition effect, and then use the data processing module to combine the shooting data and the data scanned by the radar to improve the accuracy of target recognition. When the target recognized by the shooting data is distorted, the data processing module transmits the shooting instruction to the data control module, the data control module controls the high-precision camera to shoot the target, and transmits the target data to the data processing module. The processing unit combines the shot image with the distorted image to improve the recognition accuracy of the car to the road obstacles during the automatic driving process, thereby improving the recognition effect of the entire system;
[0030] 2. The present invention uses a judgment unit to judge the simulated virtual three-dimensional model, and the deep learning module outputs optimization scheme data according to the judgment data, and uses the optimization instruction unit to adjust the optimization scheme, thereby improving the comfort of people in the car. The results in the convolutional neural network are continuously improved through the optimization scheme, and then the handling method of the vehicle in an emergency is obtained, so that the self-driving car can take the best avoidance result for property and people in an emergency.
[0031] 3. The present invention communicates with the server through the cloud. When the cloud recognizes that the reward value data in the optimization instruction unit is high, the server converts the optimization solution data into a data packet and transmits the data packet to the convolutional neural network, thereby continuously improving the output result of the convolutional neural network and optimizing the output effect of the convolutional neural network driving instruction. BRIEF DESCRIPTION OF THE DRAWINGS
[0032] Figure 1 This is a flow chart of the terminal equipment collection in an embodiment of the present invention. DETAILED DESCRIPTION
[0033] The specific embodiments of the present invention are further described below in conjunction with the accompanying drawings. It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.
[0034] The present invention provides a ring-shaped information collection terminal device for an autonomous driving vehicle, wherein the collection terminal device communicates with a vehicle driving system, and the collection terminal device includes a data collection module, a data processing module, a data control module, a data matching module, a deep learning module, a data transmission module, and a cloud;
[0035] The data acquisition module includes an environment acquisition unit and a data acquisition unit. The data acquisition unit acquires environment data around the autonomous driving vehicle through sensors. The environment data acquired by the data acquisition unit includes radar data and shooting data, and transmits the data to the data processing module. The environment acquisition unit is used to acquire environment data around the data acquisition unit. The environment data acquired by the environment acquisition unit includes light data and weather data, and transmits the data to the data processing module.
[0036] The data processing module includes a processing unit and an abnormal unit. After receiving the data from the data acquisition unit, the processing unit identifies the scanned target according to the radar data, and returns the target data of the inaccurate scanning position to the data control module. After receiving the environmental data transmitted by the environmental acquisition unit, the abnormal unit processes the light data and weather data collected by the environmental acquisition unit, and then generates a corresponding control instruction according to the processing result, and then transmits the control instruction to the data control module.
[0037] The data control module is used to adjust the start and stop of the sensor in the data acquisition unit, thereby adjusting the acquisition mode of the data acquisition unit, and converting the acquisition mode into instruction data to be transmitted to the data acquisition module;
[0038] The data matching module is used to match the identified target data with the data in the acquisition terminal device;
[0039] The deep learning module is used to determine the safety level between the vehicle motion state and the target data;
[0040] The data transmission module is used to transmit the driving instruction data to the automobile driving system;
[0041] The cloud is used to receive data around the car while driving;
[0042] The data acquisition unit includes a laser radar, a 360 camera and a high-precision camera, and uses the laser radar and the 360 camera to identify roads and obstacles around the car, and transmits the identified data to the data processing module. The processing unit is used to identify the collected road condition data, and transmits the target position data that is not accurately identified to the data control module. The data control module controls the high-precision camera to shoot the unidentified target, and transmits the photographed target data to the processing unit. The processing unit is used to match the photographed target with the inaccurately identified target, and identify the target again.
[0043] In one embodiment of the present invention: the data matching module includes a data storage unit, a storage matching unit and a judgment unit, the data storage unit is used to store virtual model data associated with road conditions, the storage matching unit matches the virtual model data in the data storage unit according to the target data identified by the data processing module, and transmits the successfully matched data to the judgment unit;
[0044] The deep learning module includes an optimization instruction unit and a convolutional neural network. The convolutional neural network is used to determine the safety level of the current car in the automatic driving situation, and determine the driving state of the car according to the safety level, and generate corresponding driving instructions, and transmit the driving instructions to the optimization instruction unit, the data transmission module and the cloud;
[0045] After receiving the environmental data, collected data and data from each module, the cloud generates and stores corresponding logs.
[0046] In one embodiment of the present invention: the judgment unit simulates the driving data of the car when driving according to the matching virtual model, and judges the current driving state of the car according to the current driving data of the car, and transmits the judgment data to the deep learning module, the optimization instruction unit is used to optimize the driving state of the car, and obtain the best optimization method, and transmit the best optimization method data to the convolutional neural network, the environment acquisition unit includes a light sensor and a meteorological sensor, the light sensor is used to obtain the light intensity when the 360 camera and the high-precision camera are shooting, and the light intensity data is transmitted to the abnormal unit, the abnormal unit adjusts the angle of the camera according to the light intensity and fills the image with light to ensure the accuracy of the 360 camera and the high-precision camera in shooting the target.
[0047] In one embodiment of the present invention: the processing unit combines radar data, 360 camera and high-precision camera, adjusts the high-precision camera to lock the target through the data processing module, and uses the 360 camera and high-precision camera to collect data of different angles of the target. The abnormal unit selects different data collection methods according to different weather conditions, and transmits the collection method data to the data control module. The data control module adjusts the number of sensors used according to the data collection method. When foggy weather and heavy rain affect the road view, the data control module transmits adjustment instructions to the data collection module, and enables laser radar, infrared camera, thermal camera and millimeter wave radar to collect road condition information in real time, to ensure that the car can avoid obstacles during driving and avoid collisions when the field of vision is blocked.
[0048] In one embodiment of the present invention: after the data processing module recognizes the target data, the data matching module matches the target data with the target virtual three-dimensional model in the data storage unit, and assembles the matched road condition model and the target model into a virtual three-dimensional model. The judgment unit judges the danger level of the car driving according to the virtual three-dimensional model, and the danger level is divided into three levels: low, medium and high. When the judgment unit is a low danger level, the optimization instruction unit intervenes, and after the convolutional neural network generates an automatic driving instruction, the driving instruction is optimized. When the judgment unit is a medium danger level and a high danger level, the optimization instruction unit does not intervene in the automatic driving instruction of the convolutional neural network. The driving instruction generated by the convolutional neural network is transmitted to the car driving system by the data transmission module, and then the optimization instruction unit generates optimization data according to the driving state, and transmits the optimization data to the convolutional neural network.
[0049] In one embodiment of the present invention, the optimization instruction unit determines the reward value that needs to be adjusted in the low-risk level state by setting a reward formula, and the specific reward formula is as follows:
[0050] X zg =f1·Z i +f2·Q i +f3·W i , (f1+f2+f3=1)
[0051] Among them, X zg represents the reward value for optimizing driving instructions, Z i Indicates the braking force, Q i Indicates the driving stability value of the car, W i represents the comfort value of the person in the car, f1, f2 and f3 represent Z i , Q i and W iThe weight coefficient of the optimized instruction unit is used to obtain the reward value that needs to be updated by the convolutional neural network when the danger level is at the medium and high danger levels. The specific reward formula is:
[0052] X zf =f5·Z i +f6·Q i +f7·W u +f8·E u , (f5+f6+f7+f8=1)
[0053] Among them, X zf Represents the reward value of updating the convolutional neural network, W u Indicates the car loss ratio, E u represents the personnel safety value, f5, f6, f7 and f8 represent Z i , Q i , W u and E u The weight coefficient is used to improve the output of the convolutional neural network using the reward formula, thereby optimizing the output of the convolutional neural network for medium and high risk levels.
[0054] The output results of the convolutional neural network are continuously improved through the above formula to ensure that the convolutional neural network can output better driving command effects when facing different target states.
[0055] In one embodiment of the present invention: a simulation calculation unit is set in the deep learning module, and the simulation calculation unit is used to obtain a virtual three-dimensional model and simulate the driving state of the car under the physical state, to ensure the feasibility of the current optimization instruction, and to transmit the feasible driving instruction data to the convolutional neural network.
[0056] In one embodiment of the present invention: a hot patch is set in the convolutional neural network. After receiving the data of the self-driving car in the cloud, the data with a reward value higher than 95% in the optimization plan is transmitted to the main server. The main server generates hot patch data according to the optimization plan and updates the convolutional neural network in the collection terminal device in the networked state.
[0057] In one embodiment of the present invention: after receiving the driving log of the car, the cloud generates a visual list of the driving log, which is convenient for personnel to connect to the cloud through the APP to retrieve the car driving data.
[0058] Example 1, please refer to the attached Figure 1The environmental factors of the car during driving are monitored through the environmental collection unit, and the environmental factor data is transmitted to the data processing module. The data processing module generates a corresponding collection strategy based on the environmental factor data, and transmits the collection strategy data to the data control module. The data control module is used to adjust the sensors of the data collection module, and collect environmental data during vehicle driving by turning off and starting different sensors.
[0059] Example 2, please refer to the attached Figure 1 , the target data around the car is matched through the data matching module, the storage matching unit matches the target data with the virtual three-dimensional model data, and the virtual three-dimensional model is transmitted to the simulation calculation unit, and the simulation calculation unit is used to calculate the driving state of the car.
[0060] Example 3, please refer to the attached Figure 1 , through the setting of thermal cameras, infrared cameras, lidar and millimeter-wave radar, it ensures that the car can still avoid obstacles in time under low visibility conditions.
[0061] Specifically, through the setting of the environment acquisition unit, the accuracy of the data acquisition unit in collecting the environmental data around the car can be guaranteed. The environmental data is used to correct the data collected by the sensor in the data acquisition unit to ensure the data acquisition effect. Then, the data processing module is used to combine the shooting data and the data scanned by the radar to improve the accuracy of target recognition. When the target recognized by the shooting data is distorted, the data processing module transmits the shooting instruction to the data control module. The data control module controls the high-precision camera to shoot the target and transmits the target data to the data processing module. The processing unit combines the shot image with the distorted image to improve the accuracy of the car's recognition of road obstacles during automatic driving, thereby improving the recognition effect of the entire system.
[0062] Specifically, the simulated virtual three-dimensional model is judged by the judgment unit, and the deep learning module outputs the optimization plan data according to the judgment data, and the optimization instruction unit is used to adjust the optimization plan, thereby improving the comfort of people in the car. The results in the convolutional neural network are continuously improved through the optimization plan, and then the handling method of the vehicle in an emergency is obtained, so that the autonomous driving car can take the best avoidance result for property and people in an emergency.
[0063] Specifically, through the communication between the cloud and the server, when the cloud recognizes that the reward value data in the optimization instruction unit is high, the server converts the optimization plan data into a data packet and transmits the data packet to the convolutional neural network, thereby continuously improving the output results of the convolutional neural network and optimizing the output effect of the convolutional neural network driving instructions.
[0064] Working principle:
[0065] The data acquisition module collects environmental data around the car and transmits the environmental data to the data processing module. The data processing module and the data control module cooperate with each other to ensure the accuracy of the data acquisition module. The deep learning module optimizes and simulates the driving instructions generated by the data acquisition module to ensure the safety of the car during automatic driving. At this point, the entire workflow is completed.
[0066] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.
[0067] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various changes and variations. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
[0068] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.
Claims
1. A ring information collection terminal device for autonomous driving vehicles, Features: The acquisition terminal device communicates with the automobile driving system, and the acquisition terminal device includes a data acquisition module, a data processing module, a data control module, a data matching module, a deep learning module, a data transmission module and a cloud; The data acquisition module collects environmental data around the autonomous driving vehicle, including radar data and shooting data; The data processing module identifies the target according to the radar data and returns the target data of the inaccurate position to the data control module; The data control module adjusts the start and stop of the sensor in the data acquisition unit, converts the acquisition mode into instruction data and transmits it to the data acquisition module; The data matching module matches the identified target data with the data in the acquisition terminal device; The deep learning module determines the safety level between the vehicle motion state and the target data; The data transmission module transmits the driving instruction data to the automobile driving system; The cloud receives data about the surroundings of the car while driving; The data acquisition unit includes a laser radar, a 360 camera and a high-precision camera. The laser radar and the 360 camera are used to identify roads and obstacles around the car, and the identified data are transmitted to the data processing module. The processing unit is used to identify the collected road condition data, and the target position data that is not accurately identified is transmitted to the data control module. The data control module controls the high-precision camera to shoot the unidentified target, and transmits the photographed target data to the processing unit. The processing unit is used to match the photographed target with the inaccurately identified target, and identify the target again.
2. The self-driving car loop information collection terminal device according to claim 1, Features: The data matching module includes a data storage unit, a storage matching unit and a judgment unit. The data storage unit is used to store virtual model data associated with road conditions. The storage matching unit matches the virtual model data in the data storage unit according to the target data identified by the data processing module, and transmits the successfully matched data to the judgment unit. The deep learning module includes an optimization instruction unit and a convolutional neural network. The convolutional neural network is used to determine the safety level of the current car in the automatic driving situation, and determine the driving state of the car according to the safety level, and generate corresponding driving instructions, and transmit the driving instructions to the optimization instruction unit, the data transmission module and the cloud; After receiving the environmental data, collected data and data from each module, the cloud generates and stores corresponding logs.
3. The self-driving car loop information collection terminal device according to claim 1, Features: The judgment unit simulates the driving data of the car according to the matching virtual model, judges the current driving state of the car according to the current driving data of the car, and transmits the judgment data to the deep learning module. The optimization instruction unit is used to optimize the driving state of the car, obtain the optimization method, and transmit the optimization method data to the convolutional neural network. The environment acquisition unit includes a light sensor and a meteorological sensor. The light sensor is used to obtain the light intensity when the 360 camera and the high-precision camera are shooting, and the light intensity data is transmitted to the abnormal unit. The abnormal unit adjusts the angle of the camera according to the light intensity and fills the image with light to ensure the accuracy of the 360 camera and the high-precision camera in shooting the target.
4. The self-driving car loop information collection terminal device according to claim 1, Features: The processing unit combines radar data, 360 camera and high-precision camera, adjusts the high-precision camera to lock the target through the data processing module, and uses the 360 camera and high-precision camera to collect data of different angles of the target. The abnormal unit selects different data collection methods according to different weather conditions, and transmits the collection method data to the data control module. The data control module adjusts the number of sensors used according to the data collection method. When foggy weather and heavy rain affect the road view, the data control module transmits adjustment instructions to the data collection module, and enables laser radar, infrared camera, thermal camera and millimeter wave radar to collect road condition information in real time, ensuring that the car can avoid obstacles during driving and avoiding collisions when the field of vision is blocked.
5. The self-driving car loop information collection terminal device according to claim 2, Features: After the data processing module recognizes the target data, the data matching module matches the target data with the target virtual three-dimensional model in the data storage unit, and assembles the matched road condition model and target model into a virtual three-dimensional model. The judgment unit judges the danger level of the car driving according to the virtual three-dimensional model, and the danger level is divided into three levels: low, medium and high. When the judgment unit is a low danger level, the optimization instruction unit intervenes, and after the convolutional neural network generates an automatic driving instruction, the driving instruction is optimized. When the judgment unit is a medium danger level or a high danger level, the optimization instruction unit does not intervene in the automatic driving instruction of the convolutional neural network. The driving instruction generated by the convolutional neural network is transmitted to the car driving system by the data transmission module, and then the optimization instruction unit generates optimization data according to the driving state, and transmits the optimization data to the convolutional neural network.
6. The self-driving car loop information collection terminal device according to claim 5, Features: The optimization instruction unit determines the reward value that needs to be adjusted under the low risk level state through the set reward formula. The specific reward formula is as follows: X zg =f 1 ·Z i +f 2 ·Q i +f 3 ·W i ,(f 1 +f 2 +f 3 =1) Among them, X zg represents the reward value for optimizing driving instructions, Z i Indicates the braking force, Q i Indicates the driving stability value of the car, W i Indicates the comfort value of people in the car, f 1 、f 2 and f 3 Respectively represent Z i , Q i and W i The weight coefficient of the optimization instruction unit is that when the danger level is at the medium danger level and the high danger level, the reward value that the convolutional neural network needs to update is obtained through the reward formula. The specific reward formula is: X zf =f 5 ·Z i +f 6 ·Q i +f 7 ·W u +f 8 ·E u ,(f 5 +f 6 +f 7 +f 8 =1) Among them, X zf Represents the reward value of updating the convolutional neural network, W u Indicates the car loss ratio, E u represents the personnel safety value, f 5 、f 6 、f 7 and f 8 Respectively represent Z i , Q i , W u and E u The weight coefficient is used to improve the output of the convolutional neural network using the reward formula, thereby optimizing the output of the convolutional neural network for medium and high risk levels.
7. The self-driving car loop information collection terminal device according to claim 2, Features: A simulation calculation unit is provided in the deep learning module, and the simulation calculation unit is used to obtain a virtual three-dimensional model and simulate the driving state of the car under the physical state, so as to ensure the feasibility of the current optimization instruction and transmit the feasible driving instruction data to the convolutional neural network.
8. The self-driving car loop information collection terminal device according to claim 2, Features: A hot patch is set in the convolutional neural network. After receiving the data of the self-driving car in the cloud, the data with a reward value higher than 95% in the optimization plan is transmitted to the main server. The main server generates hot patch data according to the optimization plan and updates the convolutional neural network in the collection terminal device in the networked state.
9. The self-driving car loop information collection terminal device according to claim 2, Features: After receiving the driving log of the car, the cloud generates a visual list of the driving log, which is convenient for personnel to connect to the cloud through the APP to retrieve the car driving data.
