Vehicle collision prediction system and method based on vehicle-mounted camera
The system uses vehicle-mounted cameras to process images for collision risk prediction, addressing data acquisition and processing challenges, thereby enhancing collision risk assessment and reducing collision likelihood.
Patent Information
- Application Number
- CN202510150550.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-11
- Publication Date
- 2025-07-15
AI Technical Summary
Existing systems lack efficient and accurate methods for data acquisition and processing to construct risk prediction models for vehicle collisions, necessitating improved data handling and risk modeling to enable timely vehicle braking measures.
A system utilizing vehicle-mounted cameras to capture and process road, environmental, and vehicle condition images, generating feature driving data for risk prediction models, which then generate collision risk probabilities and trigger appropriate braking measures.
Enables real-time collision risk prediction and reduced collision probability through efficient data processing and timely braking interventions.
Smart Images
Figure CN120308108A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of safe driving, and specifically to an automobile collision prediction system and method based on in-vehicle cameras. Background Art
[0002] Automobile collision prediction technology is an important technology related to automobile safety, aiming to identify potential collision risks of the current vehicle by arranging different advanced in-vehicle cameras, data processing methods, and collision risk prediction algorithms, and taking corresponding preventive measures to reduce the occurrence of traffic accidents and mitigate the damage caused by collisions to the vehicle.
[0003] How to efficiently and accurately obtain all the initial data related to automobile collision risk prediction, and process the obtained initial data to obtain characteristic modeling data for measuring the collision risk probability, accurately construct the corresponding risk prediction model, and then effectively predict the collision risk probability of the vehicle and timely execute the corresponding braking measures of the vehicle is an urgent problem to be solved and improved in the current field. Summary of the Invention
[0004] In order to solve the above problems, the purpose of the present invention is to provide an automobile collision prediction system and method based on in-vehicle cameras.
[0005] The purpose of the present invention can be achieved through the following technical solutions: An automobile collision prediction system based on in-vehicle cameras includes a warning center, and the warning center is communicatively connected to an in-vehicle camera module, an image processing module, a risk prediction module, and a braking control module; The in-vehicle camera module is used to control in-vehicle cameras arranged at different parts of the vehicle in real time, and obtain road images, environmental images, and vehicle condition images through the in-vehicle cameras at different parts in real time, and then merge them into an image set and transmit it to the image processing module; The image processing module is used to perform image enhancement, edge information processing, and target information detection on the image set, and then generate corresponding characteristic driving data; The risk prediction module is used to construct a corresponding risk prediction model according to the characteristic driving data, predict the collision risk of the current vehicle through the risk prediction model, and then generate the corresponding collision risk probability of the vehicle; The warning center generates corresponding different warning intervention instructions according to the collision risk probability of the vehicle, and sends the warning intervention instructions to the braking control module; The braking control module executes corresponding vehicle braking control measures according to different warning intervention instructions.
[0006] Further, the process of controlling in-vehicle cameras arranged at different parts of the vehicle in real time includes: The vehicle is divided into different parts, and the types of parts include the front area part, the rear area part, the left area part, and the right area part. Corresponding control type identifiers are bound to different parts, and the control type identifiers include Front, Rear, Left, and Right; The control type identifier Front is used to control the front area part of the vehicle in real time; The control type identifier Rear is used to control the rear area part of the vehicle in real time; The control type identifier Left is used to control the left area part of the vehicle in real time; The control type identifier Right is used to control the right area part of the vehicle in real time; The types of in-vehicle cameras include front cameras, rear cameras, and side cameras. A front camera is arranged in the front area part of the vehicle, a rear camera is arranged in the rear area part of the vehicle, and side cameras are arranged in the left area part and the right area part of the vehicle. The real-time control of the in-vehicle cameras includes imaging area control and imaging parameter control. The real-time control of different parts of the vehicle is carried out through the corresponding control type identifiers of each part.
[0007] Furthermore, the process of obtaining road images, environmental images, and vehicle condition images in real time through the in-vehicle cameras of different parts and then merging them into an image set includes: Obtain the data reading permission of the in-vehicle processing terminal set inside the vehicle, and then operate the in-vehicle processing terminal. The in-vehicle processing terminal establishes real-time communication permissions with the in-vehicle cameras of different parts of the vehicle and obtains different types of images captured by the in-vehicle cameras of different parts. The types of images include road images, environmental images, and vehicle condition images. Among them, road images are divided into front road images, rear road images, and side road images; Perform the fusion and stitching of the front road images, rear road images, and side road images, and then generate the corresponding panoramic road scene images. Merge the panoramic road scene images, environmental images, and vehicle condition images into an image set and transmit the image set to the image processing module.
[0008] Furthermore, the process of performing image enhancement, edge information processing, and target information detection on the image set and then generating the corresponding characteristic driving data includes: Take the panoramic road scene images, environmental images, and vehicle condition images in the image set as target objects in turn, obtain the color block brightness, color block contrast, and color block sharpness of each pixel block of the target object, set the comparison thresholds corresponding to the color block brightness, color block contrast, and color block sharpness respectively. When the color block brightness, color block contrast, and color block sharpness are less than their respective comparison thresholds, perform corresponding brightness increase, contrast increase, and sharpness increase, and then perform image enhancement on each target object; Set up an edge image registration library to store several edge - stitched images. After the image enhancement of each target object in the image set is completed, the edge information of the target object is processed to obtain the corner - area images of each target object, and the image defect degree of each corner - area image is obtained. Denote the image defect degree as CQ, set the storage threshold and denote it as RQ. When CQ≥RQ, mark the corner - area image corresponding to the current image defect degree as a defective corner image, and import the defective corner image into the edge image registration library. Then, match the edge - stitched image corresponding to the current defective corner image, merge the defective corner image and the edge - stitched image, and thus generate the corresponding complete edge image for each target object; Import the image set after image enhancement and edge information processing into a preset target information extraction program. Set the extraction keywords of the target information extraction program. The extraction keywords include road - feature extraction keywords, environment - feature extraction keywords, and vehicle - condition - feature extraction keywords. The target information extraction program is provided with a semantic instance segmentation area and a feature information extraction area. In the semantic instance segmentation area, the image set is segmented into several sub - object instances corresponding to different types; Perform semantic understanding of different - type sub - object instances, and thus obtain the semantic keywords of each type of sub - object instance. In the feature information extraction area, perform matching between the extraction keywords and the semantic keywords to determine whether to generate the corresponding type of target detection information. The types of target detection information include road - detection feature information, environment - detection feature information, and vehicle - condition - detection feature information. Merge the road - detection feature information, environment - detection feature information, and vehicle - condition - detection feature information, and thus generate the corresponding characteristic driving data.
[0009] Furthermore, the process of constructing a corresponding risk prediction model based on the characteristic driving data includes: Decompose the characteristic driving data into road - detection feature information, environment - detection feature information, and vehicle - condition - detection feature information, and use them as the modeling parameter items of the risk prediction model respectively. Set up a modeling program, a model library, and a feature vector library. The model library stores several model frameworks, and the feature vector library stores the modeling vectors corresponding to several risk prediction models; Input the modeling parameter items into the modeling program, and then obtain the respective frame feature information and modeling vector information of the road detection feature information, environment detection feature information, and vehicle condition detection feature information. Aggregate all the respective frame feature information to generate a frame information set, and aggregate all the respective modeling vector information to generate a vector information set. Import the frame information set into the model library, and then match the corresponding model frame in the model library that has the highest matching degree with the frame information set, and use this model frame as the modeling frame of the risk prediction model. Import the vector information set into the feature vector library, and then match all the modeling vectors in the feature vector library that meet the preset matching conditions with the vector information set. Establish a corresponding risk prediction model based on the modeling frame and the matched all modeling vectors.
[0010] Further, the process of generating the corresponding collision risk probability of the vehicle by predicting the collision risk of the current vehicle through the risk prediction model includes: Predict the collision risk of the current vehicle through the risk prediction model. The collision risk prediction sets different collision prediction items, including road prediction items, environment prediction items, and vehicle condition prediction items. Define the safe road limit conditions corresponding to the road prediction items, the reasonable environment parameters corresponding to the environment prediction items, and the safe driving vehicle condition parameters corresponding to the vehicle condition prediction items, and input them into the risk prediction model. The risk prediction model generates the corresponding collision risk probability of the current vehicle; When all of the road prediction item does not meet the safe road limit conditions, the environment prediction item does not meet the reasonable environment parameters, and the vehicle condition prediction item does not meet the safe driving vehicle condition parameters are all established, the corresponding generated collision risk probability is 85 - 99%, marked as a first-level collision risk; When only any two of the road prediction item does not meet the safe road limit conditions, the environment prediction item does not meet the reasonable environment parameters, and the vehicle condition prediction item does not meet the safe driving vehicle condition parameters are established, the corresponding generated collision risk probability is 70 - 84%, marked as a second-level collision risk; When only any one of the road prediction item does not meet the safe road limit conditions, the environment prediction item does not meet the reasonable environment parameters, and the vehicle condition prediction item does not meet the safe driving vehicle condition parameters is established, the corresponding generated collision risk probability is 60 - 69%, marked as a third-level collision risk; When all of the road prediction item does not meet the safe road limit conditions, the environment prediction item does not meet the reasonable environment parameters, and the vehicle condition prediction item does not meet the safe driving vehicle condition parameters are not established, the corresponding generated collision risk probability is 0 - 59%, marked as a fourth-level collision risk.
[0011] Further, the process of the warning center generating corresponding different warning intervention instructions according to the collision risk probability of the vehicle and sending them to the braking control module includes: The warning center generates different warning intervention instructions according to different collision risk probabilities. The warning intervention instructions include red warning instructions, orange warning instructions, yellow warning instructions, and blue warning instructions; When the collision risk probability is 85 - 99%, the corresponding warning intervention instruction generated is a red warning instruction, with an associated instruction identifier P1; When the collision risk probability is 70 - 84%, the corresponding warning intervention instruction generated is an orange warning instruction, with an associated instruction identifier P2; When the collision risk probability is 60 - 69%, the corresponding warning intervention instruction generated is a yellow warning instruction, with an associated instruction identifier P3; When the collision risk probability is 0 - 59%, the corresponding warning intervention instruction generated is a blue warning instruction, with an associated instruction identifier P4; Send different warning intervention instructions and their corresponding instruction identifiers to the braking control module.
[0012] Furthermore, the process of performing corresponding vehicle braking control measures according to different warning intervention instructions includes: The braking control module is provided with an identity verification unit and a vehicle control unit; The identity verification unit is used to determine whether the received warning intervention instruction is legal, and then decide whether to transmit it to the vehicle control unit; The vehicle control unit is used to obtain different legal warning intervention instructions, and then perform corresponding vehicle braking control measures according to different warning intervention instructions.
[0013] Furthermore, an automotive collision prediction method for an automotive collision prediction system based on an in - vehicle camera includes the following steps: Step S1, Real - time control the in - vehicle cameras arranged at different parts of the vehicle, and obtain road images, environmental images, and vehicle condition images in real time through the in - vehicle cameras at different parts, and then merge them into an image set; Step S2, Perform image enhancement, edge information processing, and target information detection on the image set, and then generate corresponding characteristic driving data; Step S3, Construct a corresponding risk prediction model according to the characteristic driving data, perform a collision risk prediction on the current vehicle through the risk prediction model, and then generate the corresponding collision risk probability of the vehicle; Step S4, The warning center generates corresponding different warning intervention instructions according to the collision risk probability of the vehicle and sends them; Step S5, Perform corresponding vehicle braking control measures according to different warning intervention instructions.
[0014] Compared with the prior art, the beneficial effects of the present invention are: 1. The vehicle-mounted camera module arranges vehicle-mounted cameras of corresponding types at different parts of the currently moving vehicle, and obtains road images, environmental images, and vehicle condition images in real time. After merging them into an image set, it is handed over to the image processing module for processing, and then generates characteristic driving data. A risk prediction model is constructed based on the characteristic driving data, and the risk prediction model is used to predict the collision risk of the current vehicle, realizing the efficient processing of data. The constructed risk prediction model can predict and display the collision risk probability of the vehicle in real time, playing a role in alerting the corresponding driver of the vehicle.
[0015] 2. The risk prediction model is used to predict the collision risk of the current vehicle and generate the corresponding collision risk probability. The warning center generates corresponding warning intervention instructions according to the collision risk probability and sends them to the braking control module. The braking control module executes the vehicle braking control measures corresponding to different warning intervention instructions, thereby greatly reducing the probability of vehicle collisions and realizing the safe driving of the vehicle. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] Figure 1 It is a schematic diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0017] As Figure 1 shown, an automotive collision prediction system based on vehicle-mounted cameras includes a warning center, and the warning center is communicatively connected to a vehicle-mounted camera module, an image processing module, a risk prediction module, and a braking control module; The vehicle-mounted camera module is used to control in real time the vehicle-mounted cameras arranged at different parts of the vehicle, and obtain road images, environmental images, and vehicle condition images in real time through the vehicle-mounted cameras at different parts, and then merge them into an image set and transmit it to the image processing module; The image processing module is used to perform image enhancement, edge information processing, and target information detection on the image set, and then generate corresponding characteristic driving data; The risk prediction module is used to construct a corresponding risk prediction model according to the characteristic driving data, predict the collision risk of the current vehicle through the risk prediction model, and then generate the corresponding collision risk probability of the vehicle; The warning center generates corresponding different warning intervention instructions according to the collision risk probability of the vehicle and sends the warning intervention instructions to the braking control module; The braking control module executes corresponding vehicle braking control measures according to different warning intervention instructions.
[0018] It should be further noted that in the specific implementation process, the process of controlling in real time the vehicle-mounted cameras arranged at different parts of the vehicle includes: The vehicle is divided into different parts, and the types of the parts include the front area part, the rear area part, the left area part, and the right area part. Corresponding control type identifiers are bound to the different parts, and the control type identifiers include Front, Rear, Left, and Right; The corresponding relationship between the control type identifier and the type of the vehicle part is as follows: The control type identifier Front is used to control the front area part of the vehicle in real time; The control type identifier Rear is used to control the rear area part of the vehicle in real time; The control type identifier Left is used to control the left area part of the vehicle in real time; The control type identifier Right is used to control the right area part of the vehicle in real time; The types of the in-vehicle cameras include a front camera, a rear camera, and side cameras. The front camera is arranged in the front area part of the vehicle, the rear camera is arranged in the rear area part of the vehicle, and the side cameras are arranged in the left area part and the right area part of the vehicle. The real-time control of the in-vehicle cameras includes imaging area control and imaging parameter control. The real-time control of different parts of the vehicle is carried out through the control type identifiers corresponding to each part; The content of the imaging area control is as follows: Obtain the actual imaging area of the in-vehicle camera in each part, and record the imaging area of the actual imaging area as Obtain the expected panoramic imaging area of each part respectively, and record the imaging area of the expected panoramic imaging area as
[0019] When Adjust the in-vehicle camera corresponding to the vehicle part that does not meet this condition, and obtain the imaging area of the new actual imaging area of the in-vehicle camera of this part in real time, and record the imaging area of the new actual imaging area as When Stop the imaging area control of the in-vehicle camera of the current part of the vehicle. When Do not perform any operation; The content of imaging parameter control is as follows: The imaging parameters include resolution, frame rate, exposure, and lens stability. Parameter ranges corresponding to the resolution, frame rate, and exposure are set. When the resolution, frame rate, and exposure are within their respective parameter ranges, it is then continued to determine whether the lens stability meets the preset desired stability. If so, all imaging parameters are qualified and no adjustment of the imaging parameters is performed. If not, the lens stability of the in-vehicle camera itself is adjusted through the gyroscope built into the in-vehicle camera until the current lens stability meets the desired stability; when any one of the resolution, frame rate, and exposure is not within its respective parameter range, it indicates that the imaging parameters are unqualified, and then parameter adjustment is performed. The types of parameter adjustment include increasing the resolution, increasing the frame rate, and increasing the exposure, so that the resolution, frame rate, and exposure are within their respective parameter ranges.
[0020] It should be further noted that in the specific implementation process, the process of obtaining road images, environmental images, and vehicle condition images in real time through in-vehicle cameras at different positions and then merging them into an image set and transmitting them to the image processing module includes: Obtain the data reading permission of the in-vehicle processing terminal set inside the vehicle, operate the in-vehicle processing terminal through the data reading permission, and the in-vehicle processing terminal establishes real-time communication permissions with in-vehicle cameras at different positions of the vehicle, so as to obtain different types of images captured by in-vehicle cameras at different positions; The types of the images include road images, environmental images, and vehicle condition images. Among them, the road images are divided into front road images, rear road images, and side road images, the environmental images are divided into static environmental images and dynamic environmental images, and the vehicle condition images include vehicle interior images and vehicle state images; Perform the fusion and stitching of the front road image, rear road image, and side road image to generate the corresponding panoramic road scene image. The content of the fusion and stitching is: sequentially perform image correction, image registration, and image fusion on different road images, obtain the distortion degree and perspective degree of the road images, mark the road images whose distortion degree and perspective degree do not meet the preset distortion threshold and perspective threshold as images to be corrected, convert them into compliant images through geometric transformation of the images to be corrected, perform image registration on the compliant images, obtain the edge feature information of each compliant image, mark each two compliant images whose edge feature information is within the preset registration feature range as a pair of registration elements, align the edge image parts of each pair of registration elements, and generate several registered images, and perform image fusion on the several registered images to generate the final panoramic road scene image; Merge the panoramic road scene image, environmental image, and vehicle condition image into an image set and transmit the image set to the image processing module.
[0021] It should be further noted that in the specific implementation process, the process of performing image enhancement, edge information processing, and target information detection on the image set, and then generating corresponding characteristic driving data includes: Taking the panoramic road scene image, environmental image, and vehicle condition image included in the image set as target objects in sequence, and performing image enhancement on each target object. The content of image enhancement is as follows: obtaining the image specification of each target object, and determining whether the image specification meets the preset standard image specification. If so, no operation is performed. If not, the image specification of the current target object is adjusted to the standard image specification; obtaining the color block brightness, color block contrast, and color block sharpness of each pixel block of the target object, setting the comparison thresholds corresponding to the color block brightness, color block contrast, and color block sharpness respectively. When the color block brightness, color block contrast, and color block sharpness are less than their respective comparison thresholds, corresponding brightness increase, contrast increase, and sharpness increase are performed; Setting up an edge image registration library, where the edge image registration library stores several edge stitching images. After the image enhancement of each target object in the image set is completed, edge information processing of the target object is performed, and the edge information processing is carried out through an edge detection algorithm; Obtaining the corner area image of each target object, and obtaining the image incompleteness degree of each corner area image. Denote the image incompleteness degree as CQ, set the storage threshold, and denote it as RQ. When CQ≥RQ, mark the corner area image corresponding to the current image incompleteness degree as a defective corner image, and import the defective corner image into the edge image registration library, and then match the edge stitching image corresponding to the current defective corner image, and merge the defective corner image and the edge stitching image, so as to generate the corresponding complete edge image of each target object; Importing the image set after image enhancement and edge information processing into a preset target information extraction program, and setting the extraction keywords of the target information extraction program. The extraction keywords include road feature extraction keywords, environmental feature extraction keywords, and vehicle condition feature extraction keywords; The target information extraction program is provided with a semantic instance segmentation area and a feature information extraction area. In the semantic instance segmentation area, the panoramic road scene image, environmental image, and vehicle condition image are respectively segmented into several sub-object instances corresponding to different types; Among them, the type of the sub-object instance after segmentation of the panoramic road scene image is a road object instance, the type of the sub-object instance after segmentation of the environmental image is an environmental object instance, and the type of the sub-object instance after segmentation of the vehicle condition image is a vehicle condition object instance; Performing semantic understanding on the road object instance, environmental object instance, and vehicle condition object instance, and then obtaining the semantic keywords of the road object instance, environmental object instance, and vehicle condition object instance respectively. The semantic keywords include the first keyword, the second keyword, and the third keyword; Among them, the first keyword is extracted corresponding to the road object instance, the second keyword is extracted corresponding to the environmental object instance, and the third keyword is extracted corresponding to the vehicle condition object instance. In the feature information extraction area, the matching between the extracted keyword and the semantic keyword is performed to determine whether to generate the target detection information of the corresponding type; Match the first keyword with the road feature extraction keyword to obtain the first matching degree. When the first matching degree exceeds the preset first threshold, the type of the generated target detection information is the road detection feature information; otherwise, no target detection information is generated; Match the second keyword with the environmental feature extraction keyword to obtain the second matching degree. When the second matching degree exceeds the preset second threshold, the type of the generated target detection information is the environmental detection feature information; otherwise, no target detection information is generated; Match the third keyword with the vehicle condition feature extraction keyword to obtain the third matching degree. When the third matching degree exceeds the preset third threshold, the type of the generated target detection information is the vehicle condition detection feature information; otherwise, no target detection information is generated; Merge the road detection feature information, the environmental detection feature information, and the vehicle condition detection feature information, and then generate the corresponding feature driving data, and transmit the feature driving data to the risk prediction module.
[0022] It should be further noted that in the specific implementation process, the process of constructing the corresponding risk prediction model according to the feature driving data includes: After the risk prediction module obtains the feature driving data, it deconstructs the feature driving data into road detection feature information, environmental detection feature information, and vehicle condition detection feature information, and respectively uses the road detection feature information, environmental detection feature information, and vehicle condition detection feature information as the modeling parameter items of the risk prediction model; Set up the modeling program, the model library, and the feature vector library. The model library stores several model frameworks, and the feature vector library stores several modeling vectors corresponding to the risk prediction models. Input the modeling parameter items into the modeling program, and then obtain the frame feature information and the modeling vector information of the road detection feature information, environmental detection feature information, and vehicle condition detection feature information respectively, and summarize all the frame feature information of each to generate a frame information set, and summarize all the modeling vector information of each to generate a vector information set; Import the frame information set into the model library, and then match the model framework in the model library that has the highest matching degree with the frame information set, and use this model framework as the modeling framework of the risk prediction model. Import the vector information set into the feature vector library, and then match all the modeling vectors in the feature vector library that meet the preset matching conditions with the vector information set. Establish the corresponding risk prediction model according to the modeling framework and the matched all modeling vectors; It should be noted that the preset matching conditions can be changed, and the matching conditions include vector weights, the cosine similarity between feature vectors, and the defined threshold for successful matching.
[0023] It should be further noted that in the specific implementation process, the process of predicting the collision risk of the current vehicle through the risk prediction model and then generating the corresponding collision risk probability of the vehicle includes: Set the data interception period, and denote the data interception period as There is where t1 is the start time of the data interception period and t2 is the end time of the data interception period. At the time point corresponding to t1, the collision risk of the current vehicle is predicted through the risk prediction model; The collision risk prediction sets different collision prediction items. The collision prediction items include road prediction items, environmental prediction items, and vehicle condition prediction items. Define the safety road limit conditions corresponding to the road prediction items, define the reasonable environmental parameters of the environmental prediction items, and define the safe driving vehicle condition parameters corresponding to the vehicle condition prediction items, and input the above-defined safety road limit conditions, reasonable environmental parameters, and safe driving vehicle condition parameters into the risk prediction model. The risk prediction model generates the corresponding collision risk probability of the current vehicle at the time point corresponding to t2; When the road prediction item does not meet the safety road limit conditions, the environmental prediction item does not meet the reasonable environmental parameters, and the vehicle condition prediction item does not meet the safe driving vehicle condition parameters all hold, the corresponding generated collision risk probability is 85 - 99%, marked as a first-level collision risk; When the road prediction item does not meet the safety road limit conditions, the environmental prediction item does not meet the reasonable environmental parameters, and the vehicle condition prediction item does not meet the safe driving vehicle condition parameters, and only any two of them hold, the corresponding generated collision risk probability is 70 - 84%, marked as a second-level collision risk; When the road prediction item does not meet the safety road limit conditions, the environmental prediction item does not meet the reasonable environmental parameters, and the vehicle condition prediction item does not meet the safe driving vehicle condition parameters, and only any one of them holds, the corresponding generated collision risk probability is 60 - 69%, marked as a third-level collision risk; When the road prediction item does not meet the safety road limit conditions, the environmental prediction item does not meet the reasonable environmental parameters, and the vehicle condition prediction item does not meet the safe driving vehicle condition parameters all do not hold, the corresponding generated collision risk probability is 0 - 59%, marked as a fourth-level collision risk.
[0024] It should be further noted that in the specific implementation process, the process by which the warning center generates corresponding different warning intervention instructions according to the collision risk probability of the vehicle and sends the warning intervention instructions to the braking control module includes: The warning center obtains the collision risk probability after the risk prediction model predicts the collision risk of the current moving vehicle, and generates corresponding different warning intervention instructions according to different collision risk probabilities. The warning intervention instructions include red warning instructions, orange warning instructions, yellow warning instructions, and blue warning instructions; When the collision risk probability is 85-99%, that is, a first-level collision risk, the corresponding warning intervention instruction generated is a red warning instruction, associated instruction identifier P1; When the collision risk probability is 70-84%, that is, a second-level collision risk, the corresponding warning intervention instruction generated is an orange warning instruction, associated instruction identifier P2; When the collision risk probability is 60-69%, that is, a third-level collision risk, the corresponding warning intervention instruction generated is a yellow warning instruction, associated instruction identifier P3; When the collision risk probability is 0-59%, that is, a fourth-level collision risk, the corresponding warning intervention instruction generated is a blue warning instruction, associated instruction identifier P4; Send the generated different warning intervention instructions and the corresponding instruction identifiers of the warning intervention instructions to the brake control module.
[0025] It should be further noted that in the specific implementation process, the process of the brake control module executing corresponding vehicle brake control measures according to different warning intervention instructions includes: The brake control module is provided with an identity authentication unit and a vehicle control unit; The identity authentication unit is used to judge whether the received warning intervention instruction is legal, and then decide whether to transmit it to the vehicle control unit; The vehicle control unit is used to obtain different legal warning intervention instructions, and then execute corresponding vehicle brake control measures according to different warning intervention instructions; The content of judging whether the warning intervention instruction is legal is as follows: Obtain the instruction identifier associated with each warning intervention instruction, that is, instruction identifiers P1, P2, P3, and P3, and obtain the respective upload IP addresses corresponding to instruction identifiers P1, P2, P3, and P4, and set an IP address set. The IP address set records a number of legal upload IP addresses. If the upload IP address of the warning intervention instruction corresponding to the current instruction identifier is not in the IP address set, it is judged that the current warning intervention instruction is illegal. Otherwise, it is judged that the current warning intervention instruction is legal; When the warning intervention instruction is legal, the warning intervention instruction is transmitted to the vehicle control unit. When the warning intervention instruction is illegal, the warning intervention instruction is sent to a preset recycle bin, and the recycle bin destroys it; When the warning intervention instruction received by the vehicle control unit is a blue warning instruction, the vehicle braking control measures to be executed are as follows: activate the automatic driving function of the vehicle, monitor the running of the current vehicle in real time, and then receive the warning intervention instruction in real time. When the warning intervention instruction changes to a higher-level warning intervention instruction, that is, a non-blue warning instruction, switch the vehicle to manual driving; When the warning intervention instructions received by the vehicle control unit are orange warning instructions and yellow warning instructions, the vehicle braking control measures to be executed are as follows: when the vehicle is in manual driving, the vehicle control unit pre-adjusts the anti-lock braking system of the vehicle and reduces the clearance distance between the brakes and the brake discs set on the vehicle to accelerate the braking reaction speed, improve the braking effect, and synchronously activate the automatic parking system set on the vehicle; When the warning intervention instruction received by the vehicle control unit is a red warning instruction, the vehicle braking control measures to be executed are as follows: urgently execute the anti-lock function of the anti-lock braking system of the vehicle, urgently shorten the clearance distance between the vehicle brakes and the brake discs to the set minimum distance, urgently activate the parking function of the automatic parking system on the vehicle, and synchronously turn on the fault warning light set on the vehicle to warn other oncoming vehicles and pedestrians on the road; The present invention also provides a vehicle collision prediction method for a vehicle collision prediction system based on an in-vehicle camera, including the following steps: Step S1, control the in-vehicle cameras arranged at different parts of the vehicle in real time, and obtain road images, environmental images, and vehicle condition images through the in-vehicle cameras at different parts in real time, and then merge them into an image set; Step S2, perform image enhancement, edge information processing, and target information detection on the image set, and then generate corresponding characteristic driving data; Step S3, construct a corresponding risk prediction model according to the characteristic driving data, predict the collision risk of the current vehicle through the risk prediction model, and then generate the corresponding collision risk probability of the vehicle; Step S4, the warning center generates corresponding different warning intervention instructions according to the collision risk probability of the vehicle and sends them; Step S5, execute corresponding vehicle braking control measures according to different warning intervention instructions.
[0026] The above embodiments are only used to illustrate the technical method of the present invention and not to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present invention.
Claims
1. An in-vehicle camera-based vehicle collision prediction system, including a warning center, characterized in that, The warning center is communicatively connected to an in-vehicle camera module, an image processing module, a risk prediction module, and a braking control module; The in-vehicle camera module is used to control in real time the in-vehicle cameras arranged at different parts of the vehicle, and obtain road images, environmental images, and vehicle condition images in real time through the in-vehicle cameras at different parts, and then merge them into an image set and transmit it to the image processing module; The image processing module is used to perform image enhancement, edge information processing, and target information detection on the image set, and then generate corresponding characteristic driving data; The risk prediction module is used to construct a corresponding risk prediction model based on the characteristic driving data, predict the collision risk of the current vehicle through the risk prediction model, and then generate the corresponding collision risk probability of the vehicle; The warning center generates corresponding different warning intervention instructions according to the collision risk probability of the vehicle, and sends the warning intervention instructions to the braking control module; The braking control module executes corresponding vehicle braking control measures according to different warning intervention instructions.
2. The vehicle collision prediction system based on an in-vehicle camera according to claim 1, wherein The process of controlling in real time the in-vehicle cameras arranged at different parts of the vehicle includes: The vehicle is divided into different parts, and the types of parts include the front area part, the rear area part, the left area part, and the right area part. Corresponding control type identifiers are bound to different parts, and the control type identifiers include Front, Rear, Left, and Right; The control type identifier Front is used to control the front area part of the vehicle in real time; The control type identifier Rear is used to control the rear area part of the vehicle in real time; The control type identifier Left is used to control the left area part of the vehicle in real time; The control type identifier Right is used to control the right area part of the vehicle in real time; The types of the in-vehicle cameras include front cameras, rear cameras, and side cameras. A front camera is arranged in the front area part of the vehicle, a rear camera is arranged in the rear area part of the vehicle, and side cameras are arranged in the left area part and the right area part of the vehicle. The real-time control of the in-vehicle cameras includes imaging area control and imaging parameter control, and the real-time control of different parts of the vehicle is carried out through the corresponding control type identifiers of each part.
3. The vehicle collision prediction system based on in-vehicle cameras according to claim 2, wherein The process of obtaining road images, environmental images, and vehicle condition images in real time through the in-vehicle cameras at different parts and then merging them into an image set includes: Obtain the data reading permission of the in-vehicle processing terminal set inside the vehicle, and then operate the in-vehicle processing terminal. The in-vehicle processing terminal establishes real-time communication permissions with the in-vehicle cameras at different parts of the vehicle, and obtains different types of images captured by the in-vehicle cameras at different parts. The types of images include road images, environmental images, and vehicle condition images, where the road images are divided into front road images, rear road images, and side road images; Perform fusion stitching of the front road images, rear road images, and side road images to generate corresponding panoramic road scene images, merge the panoramic road scene images, environmental images, and vehicle condition images into an image set, and transmit the image set to the image processing module.
4. The vehicle collision prediction system based on an in-vehicle camera according to claim 3, wherein The process of performing image enhancement, edge information processing, and target information detection on an image set and then generating corresponding characteristic driving data includes: Taking the panoramic road scene image, environment image, and vehicle condition image in the image set as target objects in sequence, obtaining the color block brightness, color block contrast, and color block sharpness of each pixel block of the target object, setting the comparison thresholds corresponding to the color block brightness, color block contrast, and color block sharpness respectively. When the color block brightness, color block contrast, and color block sharpness are less than their respective comparison thresholds, corresponding brightness increase, contrast increase, and sharpness increase are performed, and then image enhancement is performed on each target object; Setting up an edge image registration library to store several edge stitching images. When the image enhancement of each target object in the image set is completed, edge information processing of the target object is performed, obtaining the corner area image of each target object, and obtaining the image defect degree of each corner area image. Denote the image defect degree as CQ, set the storage threshold and denote it as RQ. When CQ≥RQ, mark the corner area image corresponding to the current image defect degree as a defective corner image, and import the defective corner image into the edge image registration library, and then match the edge stitching image corresponding to the current defective corner image, and merge the defective corner image and the edge stitching image, and then generate the corresponding complete edge image of each target object; Importing the image set after image enhancement and edge information processing into a preset target information extraction program, setting the extraction keywords of the target information extraction program. The extraction keywords include road feature extraction keywords, environment feature extraction keywords, and vehicle condition feature extraction keywords. The target information extraction program is provided with a semantic instance segmentation area and a feature information extraction area. In the semantic instance segmentation area, the image set is segmented into several sub-object instances corresponding to different types; Performing semantic understanding of different types of sub-object instances, and then obtaining the semantic keywords of each type of sub-object instance. In the feature information extraction area, matching is performed between the extraction keywords and the semantic keywords to determine whether to generate the target detection information of the corresponding type. The types of target detection information include road detection feature information, environment detection feature information, and vehicle condition detection feature information. Merge the road detection feature information, environment detection feature information, and vehicle condition detection feature information, and then generate the corresponding characteristic driving data.
5. The vehicle collision prediction system based on an in-vehicle camera according to claim 4, wherein The process of constructing a corresponding risk prediction model based on the characteristic driving data includes: Decomposing the characteristic driving data into road detection feature information, environment detection feature information, and vehicle condition detection feature information, and respectively using them as the modeling parameter items of the risk prediction model. Set up a modeling program, a model library, and a feature vector library. The model library stores several model frameworks, and the feature vector library stores the modeling vectors corresponding to several risk prediction models; Input the modeling parameter items into the modeling program, and then obtain the respective frame feature information and modeling vector information of the road detection feature information, environment detection feature information, and vehicle condition detection feature information. Aggregate all the respective frame feature information to generate a frame information set, and aggregate all the respective modeling vector information to generate a vector information set. Import the frame information set into the model library, and then match the corresponding model frame in the model library that has the highest matching degree with the frame information set, and use this model frame as the modeling frame of the risk prediction model. Import the vector information set into the feature vector library, and then match all the modeling vectors in the feature vector library that meet the preset matching conditions with the vector information set. Establish a corresponding risk prediction model based on the modeling frame and the matched all modeling vectors.
6. The vehicle collision prediction system based on an in-vehicle camera according to claim 5, wherein The process of predicting the collision risk of the current vehicle through the risk prediction model and then generating the corresponding collision risk probability of the vehicle includes: Predict the collision risk of the current vehicle through the risk prediction model. Different collision prediction items are set for the collision risk prediction, including road prediction items, environment prediction items, and vehicle condition prediction items. Define the safety road limit conditions corresponding to the road prediction items, the reasonable environment parameters corresponding to the environment prediction items, and the safe driving vehicle condition parameters corresponding to the vehicle condition prediction items, and input them into the risk prediction model. The risk prediction model generates the corresponding collision risk probability of the current vehicle. When the road prediction item does not meet the safety road limit conditions, the environment prediction item does not meet the reasonable environment parameters, and the vehicle condition prediction item does not meet the safe driving vehicle condition parameters all hold, the corresponding generated collision risk probability is 85 - 99%, marked as a first-level collision risk. When the road prediction item does not meet the safety road limit conditions, the environment prediction item does not meet the reasonable environment parameters, and the vehicle condition prediction item does not meet the safe driving vehicle condition parameters only any two hold, the corresponding generated collision risk probability is 70 - 84%, marked as a second-level collision risk. When the road prediction item does not meet the safety road limit conditions, the environment prediction item does not meet the reasonable environment parameters, and the vehicle condition prediction item does not meet the safe driving vehicle condition parameters only any one holds, the corresponding generated collision risk probability is 60 - 69%, marked as a third-level collision risk. When the road prediction item does not meet the safety road limit conditions, the environment prediction item does not meet the reasonable environment parameters, and the vehicle condition prediction item does not meet the safe driving vehicle condition parameters all do not hold, the corresponding generated collision risk probability is 0 - 59%, marked as a fourth-level collision risk.
7. The vehicle collision prediction system based on an in-vehicle camera according to claim 6, wherein The process of the warning center generating corresponding different warning intervention instructions according to the collision risk probability of the vehicle and sending them to the braking control module includes: The warning center generates different warning intervention instructions according to different collision risk probabilities. The warning intervention instructions include red warning instructions, orange warning instructions, yellow warning instructions, and blue warning instructions. When the collision risk probability is 85 - 99%, the corresponding generated warning intervention instruction is a red warning instruction, associated with the instruction identifier P1. When the collision risk probability is 70 - 84%, the corresponding generated warning intervention instruction is an orange warning instruction, associated with the instruction identifier P2. When the collision risk probability is 60 - 69%, the generated warning intervention instruction is a yellow warning instruction, and the associated instruction identifier is P3; When the collision risk probability is 0 - 59%, the generated warning intervention instruction is a blue warning instruction, and the associated instruction identifier is P4; Send different warning intervention instructions and their corresponding instruction identifiers to the braking control module.
8. The vehicle collision prediction system based on an in-vehicle camera according to claim 7, characterized in that, The process of performing corresponding vehicle braking control measures according to different warning intervention instructions includes: The braking control module is provided with an identity authentication unit and a vehicle control unit; The identity authentication unit is used to judge whether the received warning intervention instruction is legal, and then decide whether to transmit it to the vehicle control unit; The vehicle control unit is used to obtain different legal warning intervention instructions, and then perform corresponding vehicle braking control measures according to different warning intervention instructions.
9. The vehicle collision prediction method of the vehicle collision prediction system based on an in-vehicle camera according to any one of claims 1-8, characterized in that, It includes the following steps: Step S1, real-time control the on-vehicle cameras arranged at different parts of the vehicle, and obtain road images, environmental images and vehicle condition images in real time through the on-vehicle cameras at different parts, and then merge them into an image set; Step S2, perform image enhancement, edge information processing and target information detection on the image set, and then generate corresponding characteristic driving data; Step S3, construct a corresponding risk prediction model according to the characteristic driving data, predict the collision risk of the current vehicle through the risk prediction model, and then generate the corresponding collision risk probability of the vehicle; Step S4, the warning center generates corresponding different warning intervention instructions according to the collision risk probability of the vehicle and sends them; Step S5, perform corresponding vehicle braking control measures according to different warning intervention instructions.