System and method for advanced driver assistance leveraging auto CNN technology

Abstract

The present disclosure pertains to a system (100) and method (500) for a revolutionary smart vehicle system that significantly enhances current Advanced Driver Assistance Systems (ADAS) includes a driver assistance unit (108) through the integration of advanced artificial intelligence and optimization techniques. The system (100) employs an innovative Auto Convolutional Neural Network (Auto CNN) architecture to achieve superior performance in critical driving tasks, including lane detection, object recognition, accident prevention, and mitigation of environmental challenges such as fog and water droplet removal. A key feature of the system (100) is its real-time processing capabilities, driven by a unique Auto CNN architecture that dynamically adapts to varying driving conditions. The adaptability ensures that the system (100) consistently delivers optimal performance and reliability, enhancing overall driver safety and experience.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the field of advanced driver assistancesystems (ADAS), specifically to systems and methods utilizing Auto ConvolutionalNeural Networks (CNNs) for real-time image processing and analysis in vehicles.BACKGROUND

[0002] Advanced Driver Assistance Systems (ADAS) have become crucial inenhancing vehicle safety and aiding the transition towards autonomous driving.These systems utilize sensors, algorithms, and machine learning to assist drivers invarious conditions, but they face challenges in adverse weather and poor lighting.Traditional lane detection methods, such as Hough Transform and edge detection,often perform poorly when lane markings are faded, obscured, or compromised byenvironmental conditions, making accurate lane interpretation difficult. Theselimitations in lane detection contribute to potential risks in maintaining vehiclepositioning, which is critical for preventing accidents.

[0003] Object detection in ADAS relies on advanced sensors, including cameras,LIDAR, and radar, with deep learning models like YOLO and R-CNN used forimage recognition. However, these systems struggle with real-time processing,especially when dealing with partially obscured or unusual objects. Collisionavoidance systems, which predict the trajectory of surrounding vehicles andobstacles, can produce false positives or fail in complex situations, underminingdriver safety. Additionally, ADAS performance under extreme weather conditionsremains limited, as even high-end vehicles equipped with heated sensors fail to copeduring severe weather, highlighting the need for more advanced environmentaladaptation.

[0004] Despite the growing integration of machine learning in ADAS, manysystems still rely on static models that cannot adapt in real-time. Adaptive learningsystems, while promising, have limited practical use in commercial vehicles.Traditional optimization algorithms used for decision-making and path planning inADAS could benefit from more advanced techniques, such as genetic algorithms,which remain underutilized. Moreover, ADAS are constrained by automotivehardware limitations, creating a trade-off between functionality and real-timeperformance. The need for more adaptive, real-time systems with advancedoptimization and robust environmental capabilities is essential to advancing vehiclesafety and facilitating the move toward fully autonomous driving.

[0005] Therefore, there is a need for an adaptive learning, advanced optimizationtechniques, and improved environmental adaptation capabilities to enhance overallvehicle safety and facilitate the evolution toward fully autonomous driving.OBJECTS OF THE PRESENT DISCLOSURE

[0006] An object of the present disclosure is to provide a system capable of realtime image processing that improves visibility and clarity in various environmentalconditions, including fog, smog, rain, and snow.

[0007] An object of the present disclosure is to implement an adaptive learningframework within the ADAS that enables continuous refinement and optimizationof image processing algorithms based on real-time data and changingenvironmental conditions.

[0008] An object of the present disclosure is to develop a robust object detectionmechanism that accurately identifies and recognizes pedestrians, vehicles, andobstacles, ensuring driver awareness and enhancing safety, even under adverseweather conditions and low-light environments.

[0009] An object of the present disclosure is to create a reliable lane detectionsystem that utilizes advanced neural network architectures to dynamically adjustparameters for accurate lane boundary detection in real-time, contributing to safelane-keeping assistance.

[0010] An object of the present disclosure is to achieve efficient real-timeprocessing of complex data from multiple sensors, minimizing latency in decisionmaking processes and enhancing overall system responsiveness.

[0011] An object of the present disclosure is to generate timely visual and auditoryalerts for drivers based on detected hazards, lane deviations, and other criticalinformation, thus improving driver engagement and response time.

[0012] An object of the present disclosure is to integrate advanced environmentaladaptation techniques that enhance the system's performance in extreme weatherconditions, ensuring consistent functionality across diverse driving scenarios.

[0013] An object of the present disclosure is to leverage state-of-the-arttechnologies, including Convolutional Neural Networks (CNNs), GeneticAlgorithms (GAs), and sensor fusion methodologies, to improve the overallefficacy and reliability of the ADAS.SUMMARY

[0014] An aspect of the present disclosure pertains to a system for real-time imageprocessing in vehicles to improve visibility and safety under varying environmentalconditions, the system may be configured to include a driver assistance unitincluding a sensor unit for capturing real-time environmental data, a CNN-basedprocessing unit for image enhancement, object detection, and an alert unit forgenerating real-time notifications; a server operatively coupled to a driverassistance unit, including one or more processors configured to process real-timedata, optimize image processing techniques, and provide feedback to enhancevehicle safety and navigation.

[0015] Furthermore, the one or more processors may be coupled to a memory in adriver assistance unit, and the memory storing executable instructions, whenexecuted by the processors, cause the system to capture a plurality of imagesthrough front and rear cameras mounted on the vehicle, under weather conditionscomprising any or a combination of fog, smog, rain, snow, and varying lighting;process the captured plurality of images utilizing a Convolutional Neural Network(CNN)-based unit, wherein the CNN-based unit is designed using an Auto-CNNtechnique and optimized through a feedback mechanism to detect and reduce theimpact of fog and smog in the captured plurality of images; dynamically adjustCNN processing parameters of the CNN-based unit based on real-timeenvironmental data to optimize image quality in foggy or smoggy conditions;implement pre-processing to isolate water droplets and apply image enhancementusing a pre-trained CNN unit to remove or reduce the impact of water droplets;detect lane boundaries by processing the captured plurality of images using anadaptive CNN architecture optimized through a Genetic Algorithm (GA) anddynamically adjust CNN parameters for varying weather conditions, generatingreal-time lane detection outputs;

[0016] Continuing further, the one or more processors may be configured togenerate visual or auditory alerts to the driver based on detected lane deviations toassist in maintaining proper lane position; recognize and identify objects, includingpedestrians, vehicles, and road obstacles, using a CNN-based architecture designedwith an Auto-CNN technique and optimized through a feedback mechanism; adjustthe CNN parameters dynamically to ensure accurate object detection in bothdaytime and nighttime conditions, as well as during adverse weather, such as rain,snow, and fog; generate immediate visual or auditory alerts to the driver upondetection of objects within a predefined proximity to enhance driver awareness andresponse time; and continuously refine and update the image processing unitthrough adaptive learning based on real-time data to improve accuracy, visibility,and reliability under changing environmental conditions.

[0017] In an aspect, the CNN-based processing unit is configured to detect varyinglevels of fog density to dynamically optimize the image processing parameters.

[0018] In another aspect, the system is configured to utilize a weather forecastingmodel to pre-emptively adjust image processing parameters before entering areasof high fog or smog concentration.

[0019] In an aspect, the pre-processing unit is configured to employ a dropletdetection technique that classifies droplet size and impact on visibility to optimizethe removal process.

[0020] In an aspect, the adaptive CNN architecture is trained using a dataset thatcomprises lane patterns under various road conditions, including highways, urbanroads, and rural paths.

[0021] In an aspect, the multi-modal sensors are configured to capture images inmultiple spectrums, comprising infrared, to enhance object detection duringnighttime or low visibility conditions.

[0022] In an aspect, the CNN architecture incorporates a fusion model combiningdata from radar and LiDAR sensors to improve object detection accuracy in adverseweather.

[0023] In an aspect, the system may be configured to categorize detected objectsbased on their size, speed, and proximity to the vehicle to prioritize alert generation.

[0024] An aspect of the present disclosure is a method for real-time imageprocessing in vehicles to improve visibility and safety under varying environmentalconditions, the method may be configured to include capturing, a plurality ofimages through front and rear cameras mounted on the vehicle, under weatherconditions including any or a combination of fog, smog, rain, snow, and varyinglighting; processing, the captured plurality of images utilizing a ConvolutionalNeural Network (CNN)-based unit, wherein the CNN-based unit is designed usingan Auto-CNN technique and optimized through a feedback mechanism to detectand reduce the impact of fog and smog in the captured plurality of images.

[0025] Furthermore, the method may be include dynamically adjusting processingparameters of the CNN-based unit based on real-time environmental data tooptimize image quality in foggy or smoggy conditions; implementing a preprocessing unit to identify and isolate water droplets in the captured image stream;applying an image enhancement technique utilizing a pre-trained CNN-based unitto remove or minimize the impact of water droplets, ensuring clarity in theprocessed images; detecting lane boundaries using an adaptive CNN architectureoptimized by a Genetic Algorithm (GA), dynamically adjusting CNN parametersbased on varying weather conditions, and generating real-time lane detectionoutputs and alerts to assist the driver in maintaining lane position; recognizing andidentifying objects using a CNN-based architecture designed with an auto-CNNtechnique, dynamically adjusting CNN parameters for accurate detection in varyingconditions, and generating immediate alerts to the driver upon detecting objectswithin a predefined proximity; and continuously refining and updating the imageprocessing model through adaptive learning based on real-time data to improveaccuracy, visibility, and reliability under changing environmental conditions.

[0026] Various objects, features, aspects, and advantages of the inventive subjectmatter will become apparent from the following detailed description of preferredembodiments, along with the accompanying drawing figures in which like numeralsrepresent like components.BRIEF DESCRIPTION OF DRAWINGS

[0027] The specifications of the present disclosure are accompanied by drawingsof the system and method to aid in a better understanding of the said disclosure.The drawings are in no way limitations of the present disclosure, rather are meantto illustrate the ideal embodiments of said disclosure.

[0028] In the figures, similar components and / or features may have the samereference label. Further, various components of the same type may be distinguishedby following the reference label with a second label that distinguishes among thesimilar components. If only the first reference label is used in the specification, thedescription is applicable to any one of the similar components having the same firstreference label irrespective of the second reference label.

[0029] FIG. 1 illustrates an exemplary representation of a block diagram for realtime image processing in vehicles, in accordance with an embodiment of the presentdisclosure.

[0030] FIG. 1A illustrates an exemplary representation of a block diagram for theadvanced driver assistance system (ADAS), in accordance with an embodiment ofthe present disclosure.

[0031] FIG. 2 illustrates an exemplary representation of a module diagram andhierarchical architecture of the system, in accordance with an embodiment of thepresent disclosure.

[0032] FIG. 3 illustrates a flowchart illustrating the method for real-time fog andsmog removal from image streams within the system, in accordance with anembodiment of the present disclosure.

[0033] FIG. 4 illustrates a flowchart outlining a method for real-time removal ofwater droplets from image streams within the system, in accordance with anembodiment of the present disclosure.

[0034] FIG. 5 illustrates a flowchart outlining a method for real-time lane detectionwithin the system, in accordance with an embodiment of the present disclosure.

[0035] FIG. 6 illustrates a flowchart outlining a method for real-time detection ofobjects, in accordance with an embodiment of the present disclosure.

[0036] FIG. 7 illustrates a flowchart that outlines the process for generating theauto-CNN architecture, in accordance with an embodiment of the presentdisclosure.

[0037] FIG. 8 illustrates an exemplary embodiment of the genetic encoding ofConvolutional Neural Network (CNN) architecture, in accordance with anembodiment of the present disclosure.

[0038] FIG. 9 illustrates a flowchart detailing a method for real-time imageprocessing in vehicles aimed at enhancing visibility and safety across diverseenvironmental conditions, in accordance with an embodiment of the presentdisclosure.DETAILED DESCRIPTION

[0039] The following is a detailed description of embodiments of the disclosuredepicted in the accompanying drawings. The embodiments are in such details as toclearly communicate the disclosure. However, the amount of detail offered is notintended to limit the anticipated variations of embodiments. On the contrary, theintention is to cover all modifications, equivalents, and alternatives falling withinthe spirit and scope of the present disclosure as defined by the appended claims.

[0040] An aspect of the present disclosure relates to the present disclosureintroduces an advanced smart vehicle system designed to significantly improve theperformance of current Advanced Driver Assistance Systems (ADAS). The systemis built on an adaptive, real-time processing architecture, which leverages an AutoConvolutional Neural Network (Auto CNN) optimized by a Genetic Algorithm.This architecture enhances critical driving functionalities, including precise lanedetection, reliable object recognition (such as pedestrians and vehicles), andeffective accident prevention measures. Furthermore, it addresses commonenvironmental challenges, such as fog, rain, and low visibility, by dynamicallyadjusting its parameters to optimize performance in real time. The system's abilityto adapt to changing driving conditions ensures heightened safety, reliability, andefficiency, making it well-suited for diverse driving environments, from clearhighways to adverse weather conditions.

[0041] FIG. 1 illustrates an exemplary representation of a block diagram for realtime image processing in vehicles, in accordance with an embodiment of the presentdisclosure.

[0042] In an exemplary embodiment, referring to FIG. 1, the proposed system 100for real-time image processing in vehicles may be configured to enhance visibilityand safety across varying environmental conditions. The system 100 can includeadvanced driver assistance systems (ADAS) (interchangeably referred to as a driverassistance unit 108, hereinafter), where a driver assistance unit 108 with sensor unit110 to capture real-time environmental data and a CNN-based unit 112 for imageenhancement, an object detection unit 114, and one or more alerts unit 116(interchangeably referred to as an alert unit 116, hereinafter).

[0043] In an exemplary embodiment, the system 100 may further include a server102, where the server 102 can include one or more processors 104 (interchangeablyreferred to as processors 104, hereinafter), coupled to a memory 106 and a driverassistance unit 108. The memory 106 stores executable instructions that, whenexecuted by the processors 104, enable the system 100 to process data, optimizeimage processing techniques, and provide feedback to enhance vehicle navigation.The system 100 captures images from front and rear cameras, processes themthrough an auto-CNN optimized for reducing fog and smog effects, isolates waterdroplets, and dynamically adjusts parameters for lane detection and objectrecognition. The system 100 can generate real-time alerts for lane deviations andobstacles, continuously refining its performance based on real-time environmentaldata to adapt to changing weather and lighting conditions.

[0044] FIG. 2 illustrates an exemplary representation of a block diagram for theadvanced driver assistance system (ADAS), in accordance with an embodiment ofthe present disclosure.

[0045] In an exemplary embodiment, referring to FIG. 1A, the present disclosuremay be configured to disclose a block diagram representing a comprehensiveAdvanced Driver Assistance System (ADAS) (interchangeably referred to as adriver assistance unit 108, hereinafter). The driver assistance unit 108 can captureone or more images via vehicle-mounted cameras 118, which are first processedthrough an image preprocessing unit 132 that adjusts the resolution to meet the inputrequirements of pre-trained models 134, 136, 138.

[0046] In an embodiment, the pre-processed images undergo further analysis todetect the presence of environmental obstructions including any or a combinationof, but not limited to, fog, water, and droplets. Upon detection, the images arepassed through a dedicated fog and water droplet removal unit 134 before beingpresented on a display device 148, such as an LCD screen 148 or a Virtual Reality(VR) headset 152. In cases where no fog or water droplets are detected, the imagesare directly displayed.

[0047] Subsequent to this, noise-free images are transmitted to a pre-trained lanedetection unit 136, with the detected lane information being visually conveyed onan LCD or VR display 148, 152. Simultaneously, the noise-free images areprocessed by a pre-trained object detection unit 138, which can process the inputsfrom additional devices such as voice assist 120 and touchpads 122. The detectedobjects are then visually displayed or communicated via auditory cues using voiceassist unit 150. The noise-free images, in conjunction with input data fromultrasonic sensors 124, may be also processed by an accident detection unit 140,which can trigger visual alerts on the LCD or LED display device 148. In the eventof imminent danger, an automatic braking system is activated to prevent collisions.

[0048] In an embodiment, the driver assistance unit 108 further processes data fromvarious vehicle sensors 128, including any or a combination of, but not limited to,speed, RPM, and Battery Management System (BMS) unit 126, which aresubsequently displayed. To ensure continuous optimization and improvement ofsystem functionality, the driver assistance unit 108 may be operatively connectedto an edge computing unit 158 for performance monitoring and unit enhancement.Additionally, the driver assistance unit 108 can integrate a GPS unit 142 to providenavigational assistance to a driver.

[0049] FIG. 2 illustrates an exemplary representation of a module diagram andhierarchical architecture of the system, in accordance with an embodiment of thepresent disclosure.

[0050] In an exemplary embodiment, referring to FIG. 2, the module diagram 200can provide a detailed overview of the next-generation smart vehicle system 100(referring to as a system 100), showcasing the modular and hierarchicalarchitecture. The system 100 may be organized into seven key layers, each withspecific responsibilities. A sensor layer 214 can collect raw data from varioussources, including any or a combination of, but not limited to, visual sensors 216e.g. cameras, non-visual sensors 218 e.g. ultrasonic sensors, and environmentalsensors 220, capturing information about the vehicle's surroundings and conditions.The data is then processed by a data preprocessing layer 226, where it is cleaned,filtered, and fused to remove noise and prepare it for further analysis. An auto CNNlayer 234 can employ an adaptive Convolutional Neural Network (CNN) to analyzethe pre-processed data, dynamically adjusting to real-time conditions like fog, rain,and lighting variations, ensuring accurate detection of lanes, objects, and potentialhazards.

[0051] In an exemplary embodiment, next, a decision making layer 244 caninterpret the processed data to make high-level decisions regarding vehicleoperations, such as lane changes, obstacle avoidance, and speed control. Thedecisions are then translated into specific commands by a control layer 252, whichcan interface directly with the vehicle's systems to execute braking, acceleration,and steering actions. A human-machine interface (HMI) layer manages interactionswith the driver, providing real-time visual and auditory alerts to enhance driverawareness and support decision-making.

[0052] In an embodiment, finally, a system management layer 202, which cancoordinate the overall operation of the system 100, overseeing the interactionsbetween all layers and continuously optimizing performance. The systemmanagement layer 202 can ensure that the system functions cohesively, makingadjustments as needed to improve efficiency and adaptability under changingenvironmental conditions. This hierarchical structure enables the system to processdata in real-time, respond dynamically to driving conditions, and enhance bothsafety and performance.

[0053] FIG. 3 illustrates a flowchart illustrating the method for real-time fog andsmog removal from image streams within the system, in accordance with anembodiment of the present disclosure.

[0054] In an exemplary embodiment, referring to FIG. 3, the method 300 for realtime fog and smog removal from image streams, tailored for vehicle-mountedcameras. At step 302 the method 300 can initiate the capture of images throughfront and rear cameras, particularly in foggy or smoggy conditions. To optimizecomputational efficiency, the system first detects the presence of fog or smog in theimages. At step 304 the method 300 can prepare raw sensor data or images forfurther analysis in systems like driver assistance unit 108. The step 304 can improvevisibility and adjust the images for compatibility with higher-level tasks, any or acombination of, but not limited to, object detection or lane detection, ensuring realtime accuracy and performance. At step 306, the method 300 involves analyzingimages captured by vehicle cameras to identify the presence of visual obstructionscaused by fog or smog. Using advanced algorithms like Convolutional NeuralNetworks (CNN), the system 100 recognizes the distinctive features of fog or smog,such as reduced contrast and haziness, and flags them for further processing toenhance image clarity and visibility.

[0055] Furthermore, at step 308, if fog or smog is identified, then the step 308 goesto at step 310, therefore at step 310 involves the images undergo processing via aConvolutional Neural Network (CNN)-based unit, which is designed using anAuto-CNN algorithm. The CNN-based unit can continuously optimized through afeedback mechanism that fine-tunes the CNN to reduce the impact of fog and smogon the images. At step 312, the method ensures that high clarity and visibility aremaintained in both daytime and nighttime conditions. Furthermore, the systemadjusts its parameters dynamically based on real-time environmental data, therebyoptimizing image quality in challenging weather conditions. At step 308, thedecision block demonstrates that fog or smog is not identified, then the step 308directly goes to step 314, where the system 100 showing processed information ona visual interface.

[0056] FIG. 4 illustrates a flowchart outlining a method for real-time removal ofwater droplets from image streams within the system, in accordance with anembodiment of the present disclosure.

[0057] In an exemplary embodiment, referring to FIG. 4, the method 400 for realtime removal of water droplets from image streams in a vehicle system. At step 402,the method 400 capturing continuous images from both front and rear camerasmounted on the vehicle. At step 404, the method 400 receiving the captured imagesas an input into a pre-processing unit, specifically designed to detect and isolatewater droplets within the image stream. At step 406, the method 400 within a preprocessing unit analyzes the image stream to locate water droplets that may obstructvisibility. At step 408, a decision block within the image processing flow for realtime removal of water droplets from vehicle camera feeds. After the system 100 hascaptured the images and analyzed them to identify potential water droplets, the step408 evaluates whether any droplets have been detected in the image stream.

[0058] If the answer is "yes" (i.e., droplets are present), the method 400 proceedsto the next steps i.e. at step 410 of isolating and removing or minimizing the dropletsusing advanced techniques like a Convolutional Neural Network (CNN) and afterat step 412, the image processing unit adapts continuously to evolving weatherconditions, ensuring that the driver's visibility remains unobstructed. If "no"droplets are detected, the system 100 bypasses the droplet removal process, and theimages are passed directly to subsequent modules for further processing or displayat step 414. The decision step helps to optimize computational resources by onlyengaging the droplet removal unit when necessary, ensuring efficient real-timeperformance.

[0059] FIG. 5 illustrates a flowchart outlining a method for real-time lane detectionwithin the system, in accordance with an embodiment of the present disclosure.

[0060] In an exemplary embodiment, referring to FIG. 5, the method 500 of realtime lane detection. At step 502, the method 500 initiates with the capture of imagesfrom front-facing cameras mounted on a vehicle. At step 504, the captured imagesare first sent to a pre-processing unit that checks for the presence of fog, smog, orwater droplets. At step 506, if such obstructions are detected, they are removedusing pre-trained Convolutional Neural Network (CNN) unit. At step 508, therefined images are then processed by an adaptive CNN architecture, which isoptimized in real-time through a Genetic Algorithm (GA) to maintain highdetection accuracy.

[0061] In an exemplary embodiment, the system 100 dynamically can adjust theCNN parameters to effectively detect lane boundaries under various weatherconditions, including rain, fog, and snow. At step 510, the method 500 ensuring thatlane detection remains accurate even in changing lighting and environmentalscenarios. At step 512, lane deviations are identified, and based on these detections,the system 100 generates real-time visual or auditory alerts to assist the driver instaying within the proper lane. The method 500 can enhance overall driving safetyand reliability, particularly in adverse conditions.

[0062] Moreover, at step 506, if no fog, smog, or droplets obstructions are detected,the system 500 can proceed to further stages at step 512 of image processing withoutapplying the corrective measures. The step 506 can ensures that the vehicle'sAdvanced Driver Assistance System (ADAS) can maintain clear visibility andreliable performance in various environmental conditions.

[0063] FIG. 6 illustrates a flowchart outlining a method for real-time detection ofobjects, in accordance with an embodiment of the present disclosure.

[0064] In an exemplary embodiment, referring to FIG. 6, the method 600 is for thereal-time detection of objects in the vicinity of a vehicle. At step 602, the method600 can commence with the capture of images via multi-modal sensors, whichencompass both front and rear cameras that are integrated into the vehicle'sstructure. At step 604, the method 600 is designed to enhance image clarity bydetecting and mitigating environmental obstructions through the pre-processingunit. At step 606, the determination unit analyzes the captured images for visualindicators of the conditions. If fog, smog, or water droplets are detected, the method600 proceeds to engage relevant image enhancement processes aimed at mitigatingtheir effects at step 608. Conversely, if no obstructions are identified, the imagesmay be passed directly to subsequent processing modules without furthermodification at step 612.

[0065] In an exemplary embodiment, at step 608, 610 the images are processedutilizing an advanced Convolutional Neural Network (CNN) architecture,specifically tailored for the recognition and identification of critical objects,including any or a combination of, but not limited to, pedestrians, other vehicles,and road obstacles. At step 612, the method 600 is engineered to perform effectivelyunder various lighting conditions, including both daytime and nighttime, anddemonstrates reliable functionality even during adverse weather conditions, such asrain, snow, and fog.

[0066] At step 614, upon detecting an object within a predefined proximity to thevehicle, the system generates immediate visual or auditory alerts directed towardthe driver. The feature is pivotal in enhancing the driver's situational awareness andresponse time. Furthermore, the object detection unit is subject to continuousrefinement through adaptive learning mechanisms, leveraging real-time data toincrementally improve its accuracy and reliability over time.

[0067] FIG. 7 illustrates a flowchart that outlines the process for generating theauto-CNN architecture, in accordance with an embodiment of the presentdisclosure.

[0068] In an exemplary embodiment, referring to FIG. 7, the method 700 leveragesConvolutional Neural Networks (CNNs) for noise reduction in image processing,particularly in challenging environments. While CNNs have demonstrated superiorefficacy in eliminating noise, the design of an optimal CNN architecture that fulfillsall specified objectives presents significant challenges. One or morehyperparameters, including any or a combination of, but not limited to, the numberof nodes, layers, activation functions, kernel counts, layer types (includingconvolutional, max pooling, and batch normalization), placement of skipconnections, and optimization methods, must be meticulously selected to achieveoptimal performance.

[0069] In an exemplary embodiment, researchers often employ both theoreticalframeworks and brute-force techniques to determine the hyperparameters, strivingfor global optima. However, the architecture's configuration and the initializationof learning parameters play crucial roles in whether the model converges on globalor merely local optima concerning the loss function.

[0070] Furthermore, utilizing a Genetic Algorithm (GA) to design these complexCNN architectures enhances the likelihood of attaining global optima. Unliketraditional optimization algorithms, evolutionary algorithms, including GA, do notimpose restrictive assumptions regarding the underlying objective functions,treating them as black-box functions. This approach enables the definition ofobjective functions without necessitating profound insight into the design space'sstructure.

[0071] The architecture of the CNN, referred to as the Phenotype, is encoded intochromosomes, known as the Genotype, comprising an array of numerical valueswhere each element corresponds to a specific hyper parameter. The mapping ofeach Phenotype to its Genotype allows for training the architecture and calculatingits fitness value.

[0072] The flowchart detailing the iterative process of optimizing the CNNarchitecture via a GA includes the following key steps:1. Initial Population: At step 702, a diverse set of random CNNarchitectures is generated, forming the initial population.2. Fitness Evaluation: At step 704, each architecture is assessed againstpredefined criteria, including performance metrics, computationalefficiency, and memory usage.3. Termination Check: At step 706, the algorithm verifies whethertermination criteria, such as a specified number of generations or asatisfactory fitness level, have been achieved.4. Genetic Algorithm Process: If termination criteria are unmet, the GAproceeds with:i. Selection: At step 716, identifying the most promisingarchitectures based on fitness scores.ii. Crossover: At step 714, combining selected architectures togenerate new, potentially superior designs.iii. Mutation: At step 712, introducing random alterations topreserve diversity and explore new architectural possibilities.iv. New Generation: At step 710, formulating a new populationderived from the offspring of the genetic operations.5. Iteration: The process loops back to the fitness evaluation step for thenew generation, continuing until termination criteria are fulfilled.6. Final Architecture: At step 708, upon meeting termination criteria, thebest-performing CNN architecture is designated as the final optimizeddesign.

[0073] The method 700 harnesses evolutionary principles to autonomouslydiscover CNN architectures adept at mitigating rain noise in vehicular cameraimagery. The iterative nature of this algorithm facilitates exploration within anexpansive design space, potentially revealing innovative and highly effectivearchitectural features that may be overlooked in conventional manual designapproaches.

[0074] FIG. 8 illustrates an exemplary embodiment of the genetic encoding ofConvolutional Neural Network (CNN) architecture, in accordance with anembodiment of the present disclosure.

[0075] In an exemplary embodiment, referring to FIG. 8, the diagram 800 candepict the encoding of a Convolutional Neural Network (CNN) architecture into agenetic representation, facilitating its use within an evolutionary optimizationframework. The encoding may be configured to include several key components,outlined as follows: a chromosome 802, depicted as a light purple box, representsthe complete CNN architecture. It functions as the top-level structure comprisingmultiple genes, with each gene encoding a specific characteristic of the neuralnetwork. Genes may be illustrated as light green boxes, each gene corresponds to aspecific characteristic or component of the CNN architecture. The diagramhighlights several critical genes, including Gene 1 - layer type encoding 814: thegene encodes the type of layer employed in the CNN, any or a combination of, butnot limited to, convolutional, pooling, or batch Normalization layers.

[0076] Gene 2 - number of filters encoding 812: the gene specifies the count offilters utilized within a convolutional layer, typically varying from 16 to 256.

[0077] Gene 3 - kernel size encoding 810: the gene determines the dimensions ofthe convolutional kernel, with common options including sizes like 1x1, 3x3, and5x5.

[0078] Gene 4 - activation function encoding 808: the gene specifies the activationfunction implemented in the network, with possible selections including ReLU,LeakyReLU, and ELU.

[0079] Gene 5 - pooling type encoding 806: the gene encodes the type of poolingoperation used in the architecture, which may include Max Pooling or AveragePooling techniques.

[0080] Gene 6 - skip connection encoding 804: the gene defines the presence andconfiguration of skip connections within the network, which facilitate the flow ofinformation between non-adjacent layers.

[0081] Encoding Details: Subgraphs illustrate the potential values for each gene,showcasing how various architectural decisions are numerically represented. Thisnumeric representation enables the genetic algorithm to manipulate these valuesthroughout the evolutionary process.

[0082] Extensibility: The inclusion of the "... More Genes" box signifies thatadditional architectural features may be encoded as necessary, highlighting theflexibility and adaptability of this representation scheme to accommodate evolvingdesign requirements.

[0083] Overall, the genetic encoding framework enhances the capability of theevolutionary optimization process by systematically representing and manipulatingthe diverse architectural characteristics of the CNN, ultimately aiding in thediscovery of optimal designs tailored to specific tasks.

[0084] FIG. 9 illustrates a flowchart detailing a method for real-time imageprocessing in vehicles aimed at enhancing visibility and safety across diverseenvironmental conditions, in accordance with an embodiment of the presentdisclosure.

[0085] In an exemplary embodiment, referring to FIG. 9, a flowchart for anexemplary method 900 for real-time image processing in vehicles. At step 902, themethod 900 can include utilizing to capture a series of images using front and rearcameras installed on a vehicle. The image capture occurs under various weatherconditions, which may include fog, smog, rain, snow, and different lightingscenarios. The goal is to gather comprehensive visual data that reflects the vehicle'senvironment in challenging conditions.

[0086] Continuing further, at step 904, the captured images are processed using aConvolutional Neural Network (CNN) designed with an Auto-CNN technique,which includes a feedback mechanism for optimization. The processing aims todetect and minimize the effects of fog and smog in the images, enhancing clarityand visibility.

[0087] Continuing further, at step 906, the method can involve modifying theprocessing settings of the Convolutional Neural Network (CNN) in real-time, basedon current environmental conditions such as fog or smog. By using data from thesurroundings, the system adjusts parameters to enhance image clarity, ensuring thatthe captured images remain clear and usable despite the adverse weather effects.

[0088] Continuing further, at step 908, the method 900 can involve using one ormore processors to execute a pre-processing unit that analyzes the captured imagestream to detect and separate water droplets present in the images. The preprocessing unit applies specific algorithms or techniques to identify the waterdroplets, effectively distinguishing them from the rest of the image content, therebyfacilitating further processing or enhancement of the images for improved visibility.

[0089] Continuing further, at step 910, the method 900 can involve utilizing a pretrained Convolutional Neural Network (CNN)-based unit to apply an imageenhancement technique aimed at addressing the presence of water dropletsidentified in the previous step. The CNN processes the captured images to eitherremove or significantly reduce the visual impact of the water droplets, thusenhancing the overall clarity and quality of the processed images. By leveraging thecapabilities of the pre-trained model, the system ensures that the resulting imagesmaintain high visibility and detail, which is essential for effective visualinterpretation and subsequent analysis.

[0090] Continuous further, at step 912, the method 900 can involve using anadaptive Convolutional Neural Network (CNN) optimized by a Genetic Algorithm(GA) to detect lane boundaries in processed images. The CNN parameters aredynamically adjusted based on real-time weather conditions to ensure accuratedetection. The system generates real-time outputs and alerts to assist the driver inmaintaining proper lane positioning, thereby enhancing driving safety.

[0091] Continuous further, at step 914, the method 900 can involve utilizing aConvolutional Neural Network (CNN) architecture developed through an autoCNN technique to recognize and identify objects in the vehicle's environment. TheCNN parameters are dynamically adjusted to ensure accurate detection of objects-such as pedestrians, vehicles, and obstacles-under various conditions, includingdifferent lighting and weather scenarios. When an object is detected within aspecified proximity to the vehicle, the system generates immediate alerts to thedriver, enhancing situational awareness and improving response time for potentialhazards.

[0092] Continuous further, at step 916, the method 900 can involve the ongoingenhancement of the image processing model by employing adaptive learningtechniques that utilize real-time data. As environmental conditions change-suchas variations in weather, lighting, or road situations-the model is refined andupdated to improve its accuracy in detecting and processing images. The continuouslearning process ensures that the system remains reliable and effective in providingclear visibility, thereby enhancing the overall performance and safety of the vehicleunder diverse operational scenarios.

[0093] In summary, the present disclosure provides a system and method forenhancing vehicle safety through advanced image processing techniques. Itincludes capturing images from front and rear cameras under variousenvironmental conditions, such as fog, smog, rain, and snow. These images areprocessed using a Convolutional Neural Network (CNN) architecture optimizedthrough an Auto-CNN technique and a feedback mechanism, enabling the systemto effectively detect and mitigate the impact of adverse weather conditions. TheCNN parameters are dynamically adjusted based on real-time environmental datato optimize image quality. Additionally, a pre-processing unit identifies andisolates water droplets, followed by an image enhancement technique to ensureclarity. The system also detects lane boundaries and recognizes objects inproximity, generating immediate alerts to assist the driver. Furthermore, the imageprocessing model is continuously refined through adaptive learning using realtime data, enhancing accuracy, visibility, and reliability in changing conditions.Overall, this method significantly contributes to improved driving safety andoperational efficiency.

[0094] While the foregoing describes various embodiments of the disclosure,other and further embodiments of the disclosure may be devised without departingfrom the basic scope thereof. The scope of the disclosure is determined by theclaims that follow. The disclosure is not limited to the described embodiments,versions or examples, which are included to enable those having ordinary skill inthe art to make and use the disclosure when combined with information andknowledge available to those having ordinary skill in the art.ADVANTAGES OF THE PRESENT DISCLOSURE

[0095] The present disclosure improves image clarity and visibility underchallenging weather conditions, such as fog, smog, rain, and snow, ensuring betterdriver awareness.

[0096] The present disclosure introduces an advanced Convolutional NeuralNetworks (CNNs) that allows for real-time image analysis, providing immediatefeedback and alerts to the driver regarding lane boundaries and nearby objects.

[0097] The present disclosure features a system dynamically adjusting processingparameters based on real-time environmental data, the system optimizes imagequality and detection accuracy, adapting to varying conditions seamlessly.

Claims

1. A system (100) for real-time image processing in vehicles to improve visibility and safety under varying environmental conditions, the system (100) comprises: a driver assistance unit (108) comprising: a sensor unit (110) for capturing real-time environmental data, a CNN-based unit (112) for image enhancement, an object detection unit (114), and an alert unit (116) for generating real-time notifications; a server (102) operatively coupled to a driver assistance unit (108), comprising: one or more processors (104) configured to process real-time data, optimize image processing techniques, and provide feedback to enhance vehicle safety and navigation; and wherein the one or more processors (104) coupled to a memory (106) in a driver assistance unit (108), the memory (106) storing executable instructions, when executed by the processors (104), cause the system (100) to: capture a plurality of images through front and rear cameras (118) mounted on the vehicle, under weather conditions comprising any or a combination of fog, smog, rain, snow, and varying lighting; process the captured plurality of images utilizing a Convolutional Neural Network (CNN)-based unit (112), wherein the CNN-based unit (112) is designed using an Auto-CNN technique and optimized through a feedback mechanism to detect and reduce the impact of fog and smog in the captured plurality of images; dynamically adjust CNN processing parameters of the CNNbased unit (112) based on real-time environmental data to optimize image quality in foggy or smoggy conditions; implement pre-processing to isolate water droplets and apply image enhancement using a pre-trained CNN-based unit (112) to remove or reduce the impact of water droplets;detect lane boundaries by processing the captured plurality of images using an adaptive CNN architecture optimized through a Genetic Algorithm (GA) and dynamically adjust CNN parameters for varying weather conditions, generating real-time lane detection outputs; generate visual or auditory alerts to the driver based on detected lane deviations to assist in maintaining proper lane position; recognize and identify objects, including pedestrians, vehicles, and road obstacles, using a CNN-based architecture designed with an auto-CNN technique and optimized through a feedback mechanism; adjust the CNN parameters dynamically to ensure accurate object detection in both daytime and nighttime conditions, as well as during adverse weather, such as rain, snow, and fog; generate immediate visual or auditory alerts to the driver upon detection of objects within a predefined proximity to enhance driver awareness and response time; and continuously refine and update the image processing unit through adaptive learning based on real-time data to improve accuracy, visibility, and reliability under changing environmental conditions.

2. The system (100) as claimed in claim 1, wherein the CNN-based unit (112) is configured to detect varying levels of fog density to dynamically optimize the image processing parameters.

3. Claim 3 is Missing.

4. The system (100) as claimed in claim 1, wherein the system (100) configured to utilize a weather forecasting model to pre-emptively adjust image processing parameters before entering areas of high fog or smog concentration.

5. The system (100) as claimed in claim 1, wherein the pre-processing unit (132) is configured to employ a droplet detection technique that classifies droplet size and impact on visibility to optimize the removal process.

6. The system (100) as claimed in claim 1, wherein the adaptive CNN architecture is trained using a dataset that comprises lane patterns under various road conditions, comprising any or a combination of, highways, urban roads, and rural paths.

7. The system (100) as claimed in claim 1, wherein the multi-modal sensors configured to capture images in multiple spectrums, comprising infrared, to enhance object detection during nighttime or low visibility conditions.

8. The system (100) as claimed in claim 1, wherein the CNN architecture incorporates a fusion model combining data from radar and LiDAR sensors to improve object detection accuracy in adverse weather.

9. The system (100) as claimed in claim 1, wherein the system (100) configured to categorize detected objects based on their size, speed, and proximity to the vehicle to prioritize alert generation.

10. A method (900) for real-time image processing in vehicles to improve visibility and safety under varying environmental conditions, the method (900) comprising: capturing (902), via one or more processors (104), a plurality of images through front and rear cameras mounted on the vehicle, under weather conditions including any or a combination of fog, smog, rain, snow, and varying lighting; processing (904), via the one or more processors (104), the captured plurality of images utilizing a Convolutional Neural Network (CNN)-based unit, wherein the CNN-based unit (112) is designed using an Auto-CNN technique and optimized through a feedback mechanism to detect and reduce the impact of fog and smog in the captured plurality of images; dynamically adjusting (906), via the one or more processors (104), processing parameters of the CNN-based unit (112) based on real-time environmental data to optimize image quality in foggy or smoggy conditions; implementing (908), via the one or more processors (104), a pre-processing unit (132) to identify and isolate water droplets in the captured image stream; applying (910), via the one or more processors (104), an image enhancement technique utilizing a pre-trained CNN-based unit (112) to remove or minimize the impact of water droplets, ensuring clarity in the processed images;detecting lane boundaries (912), via the one or more processors (104), using an adaptive CNN architecture optimized by a Genetic Algorithm (GA), dynamically adjusting CNN parameters based on varying weather conditions, and generating real-time lane detection outputs and alerts to assist the driver in maintaining lane position; recognizing and identifying objects (914), via the one or more processors (104), using a CNN-based architecture designed with an auto-CNN technique, dynamically adjusting CNN parameters for accurate detection in varying conditions, and generating immediate alerts to the driver upon detecting objects within a predefined proximity; and continuously refining and updating (916), via the one or more processors (104), an image processing model through adaptive learning based on real-time data to improve accuracy, visibility, and reliability under changing environmental conditions.