Vehicle-mounted special video processing system
By establishing an outdoor environment model and using mobile rail components, multi-directional data acquisition and prediction on the outside of the vehicle is solved, and the problem of insufficient environmental data acquisition on both sides of the vehicle in the existing technology is improved, and the safety of the vehicle is improved during parking .
Patent Information
- Application Number
- CN202410958553.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-17
- Publication Date
- 2024-11-08
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing on-board video processing system is difficult to effectively obtain environmental data on both sides of the vehicle during parking, resulting in an increase in safety hazards.
By establishing an outdoor environment model, collecting and preprocessing the data on the outside of the vehicle, using a convolutional neural network to identify different types of vehicles and road conditions, and adjusting the camera position through moving rail components, rotating components and swing drives, achieving multi-directional acquisition and pre-reach Verdict.
It reduces the probability of collision with other vehicles and people crashing, improves the safety of vehicles during parking, and reduces safety risks.
Smart Images

Figure CN118918563A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of vehicle-mounted dedicated video processing, and more specifically, to a vehicle-mounted dedicated video processing system. Background Art
[0002] Cars have entered thousands of households. The acquisition of in-vehicle videos can increase the visual range during the formal process, increase safety, and can also serve as effective evidence to provide responsibility division.
[0003] The document with the prior art publication number CN214756554U provides a vehicle-mounted intelligent video image processing device, which is composed of a main acquisition component and two auxiliary acquisition components movably installed on both sides of the main acquisition component. The main acquisition component includes a shell one and a battery one fixedly installed on one side of the shell one, a support plate is fixedly installed on the side of the shell one away from the battery one, and a camera one is fixedly installed at the center of the support plate away from the battery one side, and the camera one extends out of the shell one at one end away from the support plate, and the auxiliary acquisition component includes a shell two and a battery two fixedly installed on one side of the shell two, a through hole for the movably arranged support rod is centrally provided on the side of the top end of the shell two away from the battery two, and a connecting plate is centrally provided on the outer edge surface of the support rod, a camera two is fixedly installed on one end of the connecting plate away from the support rod, and a protective shell covering the camera two is centrally provided on the side of the shell two away from the battery two, and a mounting groove for the fixed arrangement of an arc-shaped PC board is centrally provided on the side of the protective shell away from the shell two.
[0004] Although the above-mentioned prior art solution can achieve the relevant beneficial effects through the structure of the prior art, it still has the following defects: a carrier is set up and front and rear cameras are set up to obtain the front and rear direction conditions. However, during use, the two sides of the vehicle are also a relatively important range. At the same time, it is also important to obtain the two sides during parking.
[0005] Regarding the above-mentioned related technologies, the inventor believes that in the process of setting up the vehicle-specific video, the camera can acquire images from multiple directions, and at the same time, the camera can predict the positions of different areas, so as to stay at each position for different lengths of time, thereby reducing the occurrence of safety hazards.
[0006] In view of this, we propose a vehicle-specific video processing system. Summary of the invention
[0007] 1. Technical issues to be solved
[0008] The purpose of the present application is to provide a vehicle-specific video processing system, which solves the problem in the above-mentioned background technology that a carrier is set up and front and rear cameras are set up to obtain the front and rear direction conditions. However, during use, the two sides of the vehicle are also a relatively important range. At the same time, the acquisition of both sides during parking is also an important technical problem, and the technical effect is achieved.
[0009] 2. Technical solution
[0010] The technical solution of the present application provides a vehicle-specific video processing system, including:
[0011] Collecting external environment data, including data on the environment and vehicles in the corresponding directions of each area outside the vehicle;
[0012] Label the external environment data, and mark the predicted results of the external environment data;
[0013] Preprocessing of the external environment data: image loading, label missing and missing value processing of the external environment data;
[0014] Establish and train the vehicle's external environment model, use the external environment data model for training, and identify different types of vehicles, road conditions, traffic signs, etc. by learning the features in the image;
[0015] Evaluation and optimization of the vehicle exterior environment model: evaluate the accuracy and stability of the performance of the established vehicle exterior environment model and conduct training;
[0016] Deployment and monitoring of the external environment model, integrating the trained external environment model into the vehicle system for application;
[0017] The transposition device is executed to switch another position for the position that is safe according to the prediction result.
[0018] By adopting the above technical solution and establishing an external vehicle environment model, the camera can predict the locations of different areas during the setting of the vehicle-specific video, so as to stay at each location for different lengths of time. This can prevent and reduce the probability of someone hitting me instead of me, thereby reducing the occurrence of safety hazards.
[0019] As an optional solution of the technical solution of this application document, the acquisition of external vehicle environment data includes road conditions, other vehicles, pedestrians, road conditions in different weather conditions, and other vehicle conditions.
[0020] As an optional solution to the technical solution of the present application document, the annotated external vehicle environment data marks different road conditions, other vehicles, pedestrians, road conditions in different weather conditions, other vehicle conditions, situations where traffic accidents may occur, or skidding situations, and marks situations where no accidents will occur.
[0021] As an optional solution of the technical solution of this application document, the preprocessing of the vehicle external environment data, loading the vehicle external environment data image, using the image processing library to load and process the image data, the process in Python is:
[0022] Python
[0023] import cv2#Use OpenCV library
[0024] #Define the image file path
[0025] image_path='path / to / your / image.jpg'
[0026] # Load the image via OpenCV
[0027] image=cv2.imread(image_path)
[0028] # Ensure the image was loaded successfully if image is not None:
[0029] #Print the shape of the image
[0030] print("Image shape:",image.shape)
[0031] #You can perform other image processing operations or display images, etc. else:
[0032] print("Failed to load image.")
[0033] Define the path to the image file and then use OpenCV's imread function to load the image.
[0034] Handle missing values or outliers that may exist in the image by filling, removing or other methods.
[0035] As an optional solution of the technical solution of this application document, the vehicle exterior environment model is established and trained, and a convolutional neural network architecture is selected. First:
[0036] Import the necessary libraries:
[0037] Python
[0038] import tensorflow as tffrom tensorflow.keras import layers,models
[0039] Define the model architecture:
[0040] Python
[0041] model = models.Sequential()
[0042] #Add convolutional layer and pooling layer
[0043] model.add(layers.Conv2D(32,(3,3),activation='relu',input_shape=(height,width,channels)))
[0044] model.add(layers.MaxPooling2D((2,2)))
[0045] model.add(layers.Conv2D(64,(3,3),activation='relu'))
[0046] model.add(layers.MaxPooling2D((2,2)))
[0047] model.add(layers.Conv2D(128,(3,3),activation='relu'))
[0048] model.add(layers.MaxPooling2D((2,2)))
[0049] #Add fully connected layer
[0050] model.add(layers.Flatten())
[0051] model.add(layers.Dense(128,activation='relu'))
[0052] #Output layer
[0053] model.add(layers.Dense(num_classes,activation='softmax')).
[0054] As an optional solution of the technical solution of the present application document, the vehicle exterior environment model evaluation optimization adjusts and optimizes the external environment model according to the performance evaluation results, so that, according to the performance evaluation results, it is necessary to increase or decrease the convolutional layers, pooling layers, fully connected layers, etc. in the model to adjust the complexity of the model;
[0055] Using learning rate decay or the adaptive learning rate mechanism in the optimizer, the process is as follows:
[0056] from tensorflow.keras.optimizers import Adamfromtensorflow.keras.callbacks import ReduceLROnPlateau
[0057] optimizer=Adam(learning_rate=0.001)
[0058] reduce_lr=ReduceLROnPlateau(monitor='val_loss', factor=0.2, patience=3, min_lr=0.0001)
[0059] model.compile(optimizer=optimizer,loss='categorical_crossentropy',metrics=['accuracy']).
[0060] As an optional solution of the technical solution of this application document, the transposition execution device includes
[0061] A load-bearing block, the load-bearing block is arranged on the inner wall of the vehicle shell through a movable rail assembly;
[0062] A collection and processing head, which is fixedly arranged inside the load-bearing block and has a symmetrical structure;
[0063] The load-bearing block rotates the direction of the collection and processing head through a rotating assembly, increases the collection range through a swing drive, and further adjusts the position through a telescopic drive.
[0064] 3. Beneficial effects
[0065] One or more technical solutions provided in the technical solution of this application have at least the following technical effects or advantages:
[0066] 1. This application establishes an external vehicle environment model so that during the setting of the vehicle-specific video, the camera can predict the locations of different areas, and thus stay at each location for different lengths of time. This can prevent and reduce the probability of situations where I do not hit others but others hit me, thereby reducing the occurrence of safety hazards.
[0067] 2. The present application uses the setting of the mobile rail assembly to allow the cameras on both sides to acquire images from multiple directions, and after moving half a circle, the images on both sides can be acquired simultaneously, thereby reducing the acquisition time and thus reducing the probability of accidents caused by extended time.
[0068] 3. Through the setting of the rotating component, the present application can adjust the angle of the camera during the movement, so as to effectively obtain video from multiple angles, and at the same time, timely adjustments can be made to obtain more effective images.
[0069] 4. This application uses a swing drive setting to allow the camera to swing up and down, thereby expanding the acquisition range and further improving the diversity of data.
[0070] 5. This application uses the telescopic drive setting to allow the camera to move up and down to expand its range. At the same time, the camera can be lifted when not in use. BRIEF DESCRIPTION OF THE DRAWINGS
[0071] Figure 1 A structural diagram of a vehicle-mounted dedicated video processing system disclosed in a preferred embodiment of the present application;
[0072] Figure 2 This is a schematic diagram of the overall structure of a vehicle-mounted dedicated video processing system disclosed in a preferred embodiment of the present application;
[0073] Figure 3 A bottom view of the overall structure of a vehicle-mounted dedicated video processing system disclosed in a preferred embodiment of the present application;
[0074] Figure 4 This is a structural exploded diagram of a mobile rail assembly of a vehicle-mounted dedicated video processing system disclosed in a preferred embodiment of the present application;
[0075] Figure 5 This is a structural disassembly diagram of a rotating component of a vehicle-mounted dedicated video processing system disclosed in a preferred embodiment of the present application;
[0076] Figure 6 This is a disassembled diagram of the swing drive structure of a vehicle-mounted dedicated video processing system disclosed in a preferred embodiment of the present application;
[0077] Explanation of the numbers in the figure: 1. load-bearing block; 2. vehicle shell; 3. collection and processing head; 31. track; 32. guide groove; 33. connecting plate; 34. rolling wheel; 35. chassis; 36. first motor; 301. connecting shaft; 302. first telescopic rod; 303. connecting block; 3001. machine cavity; 3002. second motor; 3003. swing block; 3100. swing shaft; 3200. swing gear; 3300. rack; 3400. second telescopic rod; 3500. rotating groove. DETAILED DESCRIPTION
[0078] The present application is further described in detail below in conjunction with the accompanying drawings.
[0079] Reference Figure 1 , the embodiment of the present application provides a vehicle-specific video processing system, including:
[0080] Collecting external environment data, including data on the environment and vehicles in the corresponding directions of each area outside the vehicle;
[0081] Label the external environment data, and mark the predicted results of the external environment data;
[0082] Preprocessing of the external environment data: image loading, label missing and missing value processing of the external environment data;
[0083] Establish and train the vehicle's external environment model, use the external environment data model for training, and identify different types of vehicles, road conditions, traffic signs, etc. by learning the features in the image;
[0084] Evaluation and optimization of the vehicle exterior environment model: evaluate the accuracy and stability of the performance of the established vehicle exterior environment model and conduct training;
[0085] Deployment and monitoring of the external environment model, integrating the trained external environment model into the vehicle system for application;
[0086] The transposition device is executed to switch another position for the position that is safe according to the prediction result.
[0087] By establishing an outdoor environment model, the camera can predict the locations of different areas during the setting of the vehicle-specific video, and stay at each location for different lengths of time. This can prevent and reduce the probability of someone hitting me instead of me, thereby reducing the occurrence of safety hazards.
[0088] Reference Figure 1 and Figure 2The embodiment of the present application provides a vehicle-specific video processing system for collecting external environment data, including road conditions, other vehicles, pedestrians, road conditions in different weather conditions, and other vehicle conditions.
[0089] Label the external environment data, mark different road conditions, other vehicles, pedestrians, road conditions in different weather conditions, other vehicle conditions, situations where traffic accidents may occur, or situations where skidding may occur, and mark situations where no accidents will occur.
[0090] Preprocess the external environment data, load the external environment data image, use the image processing library to load and process the image data, the process in Python is:
[0091] Python
[0092] import cv2#Use OpenCV library
[0093] #Define the image file path
[0094] image_path='path / to / your / image.jpg'
[0095] # Load the image via OpenCV
[0096] image=cv2.imread(image_path)
[0097] # Ensure the image was loaded successfully if image is not None:
[0098] #Print the shape height, width, and number of channels of the image
[0099] print("Image shape:",image.shape)
[0100] #You can perform other image processing operations or display images, etc. else:
[0101] print("Failed to load image.")
[0102] Define the path to the image file and then use OpenCV's imread function to load the image.
[0103] Handle missing values or outliers that may exist in the image by filling, removing or other methods.
[0104] Build and train the vehicle exterior environment model and select the convolutional neural network (CNN) architecture. First:
[0105] Import the necessary libraries:
[0106] Python
[0107] import tensorflow as tffrom tensorflow.keras import layers,models
[0108] Define the model architecture:
[0109] Python
[0110] model = models.Sequential()
[0111] #Add convolutional layer and pooling layer
[0112] model.add(layers.Conv2D(32,(3,3),activation='relu',input_shape=(height,width,channels)))
[0113] model.add(layers.MaxPooling2D((2,2)))
[0114] model.add(layers.Conv2D(64,(3,3),activation='relu'))
[0115] model.add(layers.MaxPooling2D((2,2)))
[0116] model.add(layers.Conv2D(128,(3,3),activation='relu'))
[0117] model.add(layers.MaxPooling2D((2,2)))
[0118] #Add fully connected layer
[0119] model.add(layers.Flatten())
[0120] model.add(layers.Dense(128,activation='relu'))
[0121] #Output layer
[0122] model.add(layers.Dense(num_classes,activation='softmax')).
[0123] Evaluation and optimization of the external environment model: adjust and optimize the external environment model based on the performance evaluation results. Therefore, according to the performance evaluation results, it is necessary to increase or decrease the convolutional layers, pooling layers, and fully connected layers in the model to adjust the complexity of the model.
[0124] Using learning rate decay or the adaptive learning rate mechanism in the optimizer, the process is as follows:
[0125] from tensorflow.keras.optimizers import Adamfromtensorflow.keras.callbacks import ReduceLROnPlateau
[0126] optimizer=Adam(learning_rate=0.001)
[0127] reduce_lr=ReduceLROnPlateau(monitor='val_loss', factor=0.2, patience=3, min_lr=0.0001)
[0128] model.compile(optimizer=optimizer,loss='categorical_crossentropy',metrics=['accuracy']).
[0129] Perform transposition equipment, including
[0130] A load-bearing block 1, which is arranged on the inner wall of the vehicle shell 2 through a movable rail assembly;
[0131] The collecting and processing head 3 is fixedly arranged inside the load-bearing block 1 and is arranged in a symmetrical structure;
[0132] The load-bearing block 1 rotates the direction of the collection and processing head 3 through a rotating assembly, increases the collection range through a swing drive, and further adjusts the position through a telescopic drive.
[0133] Reference Figure 2 and Figure 3 , the embodiment of the present application provides a vehicle-mounted dedicated video processing system, including a load-bearing block 1, which is arranged on the inner wall of a vehicle shell 2 through a movable rail assembly;
[0134] The collecting and processing head 3 is fixedly arranged inside the load-bearing block 1 and is arranged in a symmetrical structure;
[0135] The load-bearing block 1 rotates the direction of the collection and processing head 3 through a rotating assembly, increases the collection range through a swing drive, and further adjusts the position through a telescopic drive.
[0136] Reference Figure 4 , the embodiment of the present application provides a vehicle-mounted dedicated video processing system, the mobile rail assembly includes a rail 31, the rail 31 is arranged in a square structure, and the rail 31 is fixedly arranged on the inner wall of the vehicle shell 2;
[0137] A guide groove 32 is provided inside the track 31, and a rolling wheel 34 is fixedly arranged inside the guide groove 32. The rolling wheels 34 are arranged in a left and right structure with two, and the two rolling wheels 34 are rotatably connected through a connecting plate 33;
[0138] A chassis 35 is fixedly disposed on the outer wall of the connecting plate 33 , and a first motor 36 is disposed inside the chassis 35 . The output shaft of the first motor 36 extends through one side of the connecting plate 33 to the other side and is coaxially fixedly connected to one of the rolling wheels 34 .
[0139] The rotation of the output shaft of the first motor 36 drives one of the rolling wheels 34 to move. At this time, the two rolling wheels 34 will be connected by rolling in the guide groove 32, so that the load-bearing block 1 can move along the track of the track 31, thereby driving the two symmetrical acquisition and processing heads 3 on the load-bearing block 1 to move to acquire images around the vehicle. Through the setting of the mobile rail assembly, the cameras on both sides can acquire images from multiple directions, and after moving half a circle, the images on both sides can be acquired simultaneously, which reduces the acquisition time and thus reduces the probability of accidents caused by extended time.
[0140] Reference Figure 5 , the embodiment of the present application provides a vehicle-mounted dedicated video processing system, the telescopic drive includes a connecting shaft 301 fixedly arranged on the outer wall of the connecting plate 33, a first telescopic rod 302 is fixedly arranged on the outer end of the connecting shaft 301, and the connecting shaft 301 and the first telescopic rod 302 are arranged in an L-shaped structure;
[0141] A connecting block 303 is fixedly disposed at the other end of the first telescopic rod 302 .
[0142] The extension and retraction of the first telescopic rod 302 drives the load-bearing block 1 to move up and down, thereby expanding the range and angle of the camera. Through the setting of the telescopic drive, the camera can move up and down to expand the range. At the same time, the camera can be lifted when not in use.
[0143] Reference Figure 4 and Figure 5The embodiment of the present application provides a vehicle-specific video processing system, wherein the rotating component includes a machine cavity 3001 opened inside a connecting block 303, a second motor 3002 is arranged inside the machine cavity 3001, and an output shaft of the second motor 3002 extends through the inner wall of the machine cavity 3001 to the outside and is coaxially fixed with a swing block 3003.
[0144] The rotation of the output shaft of the second motor 3002 drives the swing block 3003 to rotate, so that the load-bearing block 1 moves in a circle, so as to adjust the angle of different positions to obtain better and more accurate image data. Through the setting of the rotating component, the angle of the camera can be adjusted during the movement, so as to effectively obtain video from multiple angles, and timely adjustments can be made to obtain more effective images.
[0145] Reference Figure 6 , the embodiment of the present application provides a vehicle-specific video processing system, the swing drive includes a swing shaft 3100 fixedly arranged on the outer wall of the load-bearing block 1, and the swing shaft 3100 is rotatably arranged inside a rotating groove 3500 opened on the outer wall of the swing block 3003;
[0146] A swing gear 3200 is fixedly arranged in the middle of the swing shaft 3100 , a rack 3300 is meshedly connected to the outer wall of the swing gear 3200 , and a second telescopic rod 3400 is fixedly arranged at one end of the rack 3300 .
[0147] The rack 3300 is driven to move by the extension and retraction of the second telescopic rod 3400, and the rack 3300 is meshed with the swing gear 3200, thereby driving the swing shaft 3100 to rotate, so that the load-bearing block 1 can swing at a certain angle. Through the setting of the swing drive, the camera can swing up and down, thereby expanding the acquisition range and further improving the diversity of data.
[0148] When the vehicle-specific video processing system is needed, first, collect the data of the external environment, collect the data of the environmental vehicles in the corresponding directions of each area outside the vehicle, and obtain the external environment data including road conditions, other vehicles, pedestrians, and road conditions in different weathers, and other vehicle conditions. At this time, mark the different road conditions, other vehicles, pedestrians, and road conditions in different weathers, other vehicle conditions, possible traffic accidents, or slipping conditions, and mark the situations where no accidents will occur. Preprocess the external environment data, load the external environment data image, and use the image processing library to load and process the image data. The process in Python is:
[0149] Python
[0150] import cv2#Use OpenCV library
[0151] #Define the image file path
[0152] image_path='path / to / your / image.jpg'
[0153] # Load the image via OpenCV
[0154] image=cv2.imread(image_path)
[0155] # Ensure the image was loaded successfully if image is not None:
[0156] #Print the shape height, width, and number of channels of the image
[0157] print("Image shape:",image.shape)
[0158] #You can perform other image processing operations or display images, etc. else:
[0159] print("Failed to load image.")
[0160] Define the path to the image file and then use OpenCV's imread function to load the image.
[0161] To process the missing values or outliers that may exist in the image, we can do so by filling, removing or other methods. We can use the external environment data model for training and select the convolutional neural network (CNN) architecture. First:
[0162] Import the necessary libraries:
[0163] Python
[0164] import tensorflow as tffrom tensorflow.keras import layers,models
[0165] Define the model architecture:
[0166] Python
[0167] model = models.Sequential()
[0168] #Add convolutional layer and pooling layer
[0169] model.add(layers.Conv2D(32,(3,3),activation='relu',input_shape=(height,width,channels)))
[0170] model.add(layers.MaxPooling2D((2,2)))
[0171] model.add(layers.Conv2D(64,(3,3),activation='relu'))
[0172] model.add(layers.MaxPooling2D((2,2)))
[0173] model.add(layers.Conv2D(128,(3,3),activation='relu'))
[0174] model.add(layers.MaxPooling2D((2,2)))
[0175] #Add fully connected layer
[0176] model.add(layers.Flatten())
[0177] model.add(layers.Dense(128,activation='relu'))
[0178] #Output layer
[0179] model.add(layers.Dense(num_classes,activation='softmax'))
[0180] Evaluate the accuracy and stability of the performance of the established vehicle exterior environment model and conduct training. According to the performance evaluation results, adjust and optimize the exterior environment model. Therefore, according to the performance evaluation results, it is necessary to increase or decrease the convolutional layers, pooling layers, fully connected layers, etc. in the model to adjust the complexity of the model.
[0181] Using learning rate decay or the adaptive learning rate mechanism in the optimizer, the process is as follows:
[0182] from tensorflow.keras.optimizers import Adamfromtensorflow.keras.callbacks import ReduceLROnPlateau
[0183] optimizer=Adam(learning_rate=0.001)
[0184] reduce_lr=ReduceLROnPlateau(monitor='val_loss', factor=0.2, patience=3, min_lr=0.0001)
[0185] model.compile(optimizer=optimizer,loss='categorical_crossentropy',metrics=['accuracy']);
[0186] The trained vehicle exterior environment model is integrated into the vehicle system for application. The output shaft of the first motor 36 rotates to drive one of the rolling wheels 34 to move. At this time, the two rolling wheels 34 will be connected by rolling in the guide groove 32, so that the load-bearing block 1 moves along the track of the track 31, thereby driving the two symmetrical acquisition and processing heads 3 on the load-bearing block 1 to move to acquire the surroundings of the vehicle. At the same time, the extension and retraction of the first telescopic rod 302 drives the load-bearing block 1 to move up and down, thereby expanding the range and angle of the camera. At the same time, the output shaft of the second motor 3002 rotates to drive the swing block 3003 to rotate, so that the load-bearing block 1 performs a circular motion, thereby adjusting the angle of different positions to obtain better and more accurate image data. At the same time, the rack 3300 is driven to move through the extension and retraction of the second telescopic rod 3400, and the rack 3300 is engaged with the swing gear 3200, thereby driving the swing shaft 3100 to rotate, so that the load-bearing block 1 swings the angle, which can expand the acquisition range and further improve the diversity of data. According to the prediction result, the safe position of the prediction result is switched to another position.
Claims
1. A vehicle-mounted dedicated video processing system, characterized in that: Include: Collecting external environment data, including data on the environment and vehicles in the corresponding directions of each area outside the vehicle; Label the external environment data, and mark the predicted results of the external environment data; Preprocessing of the external environment data: image loading, label missing and missing value processing of the external environment data; Establish and train the vehicle's external environment model, use the external environment data model for training, and identify different types of vehicles, road conditions, traffic signs, etc. by learning the features in the image; Evaluation and optimization of the vehicle exterior environment model: evaluate the accuracy and stability of the performance of the established vehicle exterior environment model and conduct training; Deployment and monitoring of the external environment model, integrating the trained external environment model into the vehicle system for application; The transposition device is executed to switch another position for the position that is safe according to the prediction result.
2. The vehicle-mounted dedicated video processing system according to claim 1, characterized in that: The acquisition of the external environment data includes road conditions, other vehicles, pedestrians, road conditions in different weather conditions, and other vehicle conditions.
3. The vehicle-mounted dedicated video processing system according to claim 1, characterized in that: The labeled external vehicle environment data marks different road conditions, other vehicles, pedestrians, road conditions in different weather conditions, other vehicle conditions, situations where traffic accidents may occur, or situations where skidding may occur, and marks situations where no accidents will occur.
4. The vehicle-mounted dedicated video processing system according to claim 1, characterized in that: The preprocessing of the vehicle external environment data is to load the image of the vehicle external environment data, and use the image processing library to load and process the image data. The process in Python is as follows: Python import cv2#Use OpenCV library #Define the image file path image_path='path / to / your / image.jpg' # Load the image via OpenCV image=cv2.imread # Ensure the image was loaded successfully if image is not None: #Print the shape of the image print #You can perform other image processing operations or display images, etc. else: print Define the path to the image file and then use OpenCV's imread function to load the image; Handle missing values or outliers that may exist in the image by filling, removing or other methods.
5. The vehicle-mounted dedicated video processing system according to claim 1, characterized in that: The above-mentioned external environment model is established and trained, and a convolutional neural network architecture is selected. First: Import the necessary libraries: Python import tensorflow as tffrom tensorflow.keras import layers,modelsDefine the model architecture: Python model = models.Sequential #Add convolutional layer and pooling layer model.add(layers.Conv2D(32,,activation='relu',input_shape= model.add(layers.MaxPooling2D( model.add(layers.Conv2D(64, model.add(layers.MaxPooling2D( model.add(layers.Conv2D(128, model.add(layers.MaxPooling2D( #Add fully connected layer model.add(layers.Flatten model.add(layers.Dense #Output layer model.add(layers.Dense.
6. The vehicle-mounted dedicated video processing system according to claim 1, characterized in that: The vehicle exterior environment model evaluation optimization adjusts and optimizes the exterior environment model according to the performance evaluation results, thereby, according to the performance evaluation results, it is necessary to increase or decrease the convolutional layers, pooling layers, fully connected layers, etc. in the model to adjust the complexity of the model; Using learning rate decay or the adaptive learning rate mechanism in the optimizer, the process is as follows: from tensorflow.keras.optimizers import Adamfromtensorflow.keras.callbacks import ReduceLROnPlateau optimizer = Adam reduce_lr=ReduceLROnPlateau model.compile.
7. The vehicle-mounted dedicated video processing system according to claim 1, characterized in that: The transposition device comprises A load-bearing block, the load-bearing block is arranged on the inner wall of the vehicle shell through a movable rail assembly; A collection and processing head, which is fixedly arranged inside the load-bearing block and has a symmetrical structure; The load-bearing block rotates the direction of the collection and processing head through a rotating assembly, increases the collection range through a swing drive, and further adjusts the position through a telescopic drive.
