SmartClaim F ppt.pptxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

"Smart claim" refers to a modern approach to managing insurance claims, often using technology like artificial intelligence (AI) and automation to streamline the process. These systems aim to improve efficiency and accuracy by automating tasks like status checking, fraud detection, and decision-making, which can lead to faster processing, reduced costs, and a better customer experience.

1. EAST WEST INSTITUTE OF TECHNOLOGY BENGALURU-560091 (Affiliated to Visvesvaraya Technological University, Belgaum, Karnataka) DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING PROJECT PRESENTATION ON “SmartClaim: Deep Learning-Driven Framework for Vehicle Damage Classification and Cost Estimation” BACHELORE OF ENGINEERING IN COMPUTER SCIENCE & ENGINEERING Under the Guidance of Dr Usha B S Professor Dept. of CSE Submitted by AAKARSHANA S G 1EW21CS002 YASHASWINI J N 1EW21CS176 SAHANA S 1EW21CS134 DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

2. LIST OF CONTENTS INTRODUCTION ABSTRACT LITERATURE SURVEY PROBLEM STATEMENT OBJECTIVES PROPOSED SYSTEM IMPLEMENTATION AND TESTING RESULTS CONCLUSION REFERENCES

3. INTRODUCTION Need. Millions of vehicular accidents occur globally, necessitating accurate and timely damage assessment. Manual damage inspections are time-consuming, prone to human error, and often influenced by bias. Vehicle rental and insurance industries require a reliable, automated system to streamline inspections and reduce delays. Fig. Traditional insurance claiming process. Dept. of CSE EWIT 3

4. 2. Historical context. Conventional inspections involve visual checks by inspectors, causing delays in insurance claims and repair cost estimations. Advances in machine learning and image recognition, particularly Convolutional Neural Networks (CNNs), have transformed automated damage detection. CNN models like VGG16 have proven effective in image classification and visual recognition tasks. Fig. Evolution of Vehicle Damage Assessment Methods. Dept. of CSE EWIT 4

5. 3. Overview. This project presents a deep learning-driven framework for vehicle damage classification and cost estimation. Uploaded vehicle images are processed to detect damage, classify severity levels, and predict repair costs based on the vehicle’s specifications. The system generates a comprehensive report, enhancing efficiency and accuracy in damage assessment. Fig. Workflow of Automated Vehicle Damage Assessment System. Dept. of CSE EWIT 5

6. ABSTRACT This project introduces a web-based system for automated car damage detection and cost estimation. It leverages the YOLO object detection model to identify damaged car parts from uploaded images. Flask is used to manage user interaction, image processing, and result display. Predefined part prices are utilized to calculate repair costs based on detected damages. The system ensures accuracy, efficiency, and user-friendliness, reducing manual effort in damage assessment. Dept. of CSE EWIT 6

7. TITLE AUTHORS YEAR OBJECTIVES ADVANTAGES DISADVANTAGES Automated Vehicle Damage Detection and Repair Cost Estimation Using Deep Learning Sunil Kumar Aithal S K Chirag Nackathaya Dhanush Poojary Gautham Bhandary Avinash Acharya 2024 To develop an automated system using deep learning to detect vehicle damages, estimate repair costs, and provide a quick, user-friendly platform for stakeholders. The automated system enhances efficiency and reliability by reducing reliance on manual inspections, offering high accuracy in damage detection through state-of-the-art deep learning models, and streamlining insurance and repair workflows. Requires high-quality images and preprocessing , with limited generalization to diverse vehicles and damages. High computational demands and reliance on fixed cost standards may limit scalability. A System for Automated Vehicle Damage Localization and Severity Estimation Using Deep Learning Yantai Ma Hiva Ghanbari Tianyuan Huang Jeremy Irvin 2024 To design an automated system that uses deep learning for vehicle damage localization and severity estimation by analyzing user-submitted photographs, generating a detailed damage report. Utilizes deep learning modules for vehicle identification, part localization, and damage localization, offering flexible, scalable solutions integrated with customer workflows. Utilizes deep learning modules for vehicle identification, part localization, and damage localization. Challenges in generalization to new vehicle models. LITE R ATURE SURVEY Dept. of CSE EWIT 7

8. Vehicle Damage Severity Estimation for Insurance Operations Using In-The-Wild Mobile Images Dimitrios Mallios Li Xiaofei Niall Mclaughlin Jesus Martinez Del Rincon Clare Galbraith Rory Garland 2023 To develop an automated system for vehicle damage severity estimation and cost assessment using multi-angle images and structured vehicle data. Increased Efficiency: Automates the vehicle damage assessment process, significantly reducing time and labor costs. Image Quality Dependency: The system's accuracy relies heavily on the quality and clarity of user-uploaded images. Limited Context Understanding: The model may struggle to detect internal or non-visible damages that are not captured in the images. Vehicle-Damage-Detection Segmentation Algorithm Based on Improved Mask RCNN Qinghui Zhang Xianing Chang Shanfeng Bian 2020 To develop an improved Mask RCNN-based vehicle damage detection algorithm that enhances average precision, detection accuracy, and masking accuracy, enabling efficient and accurate resolution of traffic accident compensation issues. Improved accuracy and efficiency in vehicle damage detection, enabling faster resolution of traffic accident compensation through optimized Mask RCNN. Limited to detecting external vehicle damage, unable to assess internal issues. Performance may decrease under challenging conditions like poor lighting or unusual angles. TITLE AUTHORS YEAR OBJECTIVES ADVANTAGES DISADVANTAGES Dept. of CSE EWIT 8

9. Vehicle Damage Classification and Fraudulent Image Detection Including Moiré Effect Using Deep Learning Umer Waqas Nimra Akram Soohwa Kim Donghun Lee Jihoon Jeon 2020 To develop a lightweight deep learning-based system that classifies vehicle damages into three categories (medium, huge, and no damage) while incorporating a hybrid approach to detect and filter fraudulent images Robust detection of fraudulent images through metadata analysis and Moiré effect detection, achieving 99% accuracy in ensuring image authenticity. 1. Limited to external cosmetic damages, lacking capability to assess internal vehicle issues such as engine or structural alignment. 2. Performance may be affected by challenging conditions like poor lighting, awkward angles TITLE AUTHORS YEAR OBJECTIVES ADVANTAGES DISADVANTAGES Dept. of CSE EWIT 9

10. PROBLEM STATEMENT To address these issues, this project aims to develop an automated system that utilizes the YOLO object detection model to identify damaged vehicle parts from uploaded images and estimate repair costs based on predefined pricing for faster, accurate, and objective evaluations, enhancing efficiency and transparency for both insurers and policyholders. Dept. of CSE EWIT 10

11. OBJECTIVES The Main objectives of the project are:  Develop a web-based platform for automated car damage detection and cost estimation. Integrate YOLO object detection to accurately identify damaged car parts from images. Enable real-time calculation of repair costs based on predefined part prices. Provide a user-friendly interface for image upload and displaying results. Ensure modular and scalable architecture for future enhancements and customizations. Enhance efficiency in damage assessment, reducing manual effort and time. Dept. of CSE EWIT 11

12. PROPOSED SYSTEM Architecture/Design. System Architecture. The system architecture follows a web-based client-server model: Client-Side: HTML templates (dashboard.html, estimate.html) for user interaction. Users can upload images of damaged cars. Server-Side: Flask server processes the uploaded image. YOLO model is used to detect car parts in the image. Damage estimation logic calculates the total cost of repairs based on detections. Dept. of CSE EWIT 12

13. Fig. Architecture of system Dept. of CSE EWIT 13

14. . Data Storage: Uploaded and processed images are stored in the static/ directory. Dept. of CSE EWIT 14

15. Dept. of CSE EWIT 15 Fig. Flowchart of architecture

16. 2. Algorithms/Mathematical Models Algorithm for Damage Estimation. Input: Uploaded image ( image_path ), predefined part prices (PART_PRICES). Processing: Pass the image through the YOLO model to detect objects. Extract detected part IDs ( class_ids ) and their counts ( class_counts ). Match detected IDs to car parts using a predefined class-to-part mapping. Multiply detected counts by their respective part prices. Output: Cost estimation for individual parts and total repair cost. Upload → Detect → Calculate → Display Results. Fig. Flowchart of algorithm. Dept. of CSE EWIT 16

17. Mathematical Model Detection Mapping: Cost calculation: Where: n: Number of detected part types. Count𝑖: Number of instances detected for part 𝑖. Price𝑖​: Price for one instance of part 𝑖. Dept. of CSE, EWIT 17

18. Dept. of CSE, EWIT 18 3. Module-wise Description: Flask Application Functionality: Routes (/, /dashboard) to manage user interaction. Renders templates (dashboard.html, estimate.html) for the user interface. Key Functions: dashboard(): Handles image uploads, processes the image, and calculates estimates. get_damage_info ( class_counts ): Maps detected parts to names and computes cost estimates. Image Processing and Detection YOLO Integration: Loads a pre-trained YOLO model (best.pt). Processes uploaded images to detect car parts and their locations.Output:Detected objects with class IDs, bounding boxes, and counts. Output: Detected objects with class IDs, bounding boxes, and counts.

19. Dept. of CSE, EWIT 19 Damage Estimation Predefined Prices: Dictionary PART_PRICES contains prices for each car part. Logic: Maps detected class IDs to car part names. Computes total cost for each part based on the count. File Management Image Handling: Uses secure_filename to handle file uploads securely. Saves uploaded and processed images in the static/ directory. Error Handling: Ensures only valid image files are uploaded. User Interface Templates: dashboard.html: Upload interface for images. estimate.html: Displays detected parts, counts, and cost estimates. Flask Flash Messages: Used to notify users of errors (e.g., invalid file type).

20. Dept. of CSE, EWIT 20 IMPLEMENTATION Dashboard

21. Dept. of CSE, EWIT 21 2. Image detection. 3. Cost Estimation.

22. Dept. of CSE, EWIT 22 4. User Interface and Results Display. Dashboard Template Rendering: Estimate Template Rendering:

23. TESTING Validate the YOLOv8-based car damage detection model for accuracy and efficiency. Test Environment: Dataset: Car Damage Detection.v2i.yolov8. Classes: Bonnet, Bumper, Dickey, Door, Fender, Light, Windshield. Training Configuration: Model: YOLOv8n (pretrained). Optimizer: Adam with cosine learning rate scheduling. Device: GPU (if available). Performance Metrics: mAP50: Measures detection accuracy at a single IoU threshold of 50%. mAP50-95: Measures detection accuracy across multiple IoU thresholds. Dept. of CSE, EWIT 23

24. USE CASES Car Damage Detection System: Input: Image of a car with visible damages (e.g., damaged bumper and windshield). Expected Output: Identification of damaged parts with bounding boxes and cost estimation based on predefined prices. Actual Output: The system accurately detected the damaged bumper and windshield, providing repair costs consistent with predefined rates. Vehicle Inspection Assistance: Input: Image of a vehicle submitted for routine inspection. Expected Output: Detection of damages, if any, and generation of a report for repair suggestions. Actual Output: The model identified minor scratches on the fender and estimated repair costs, enabling timely maintenance. Dept. of CSE EWIT 24

25. RESULTS 1. Homepage (Dashboard). Dept. of CSE, EWIT 25 Fig. User Dashboard for Image Upload and Detection.

26. Dept. of CSE, EWIT 26 2. Warning for Missing File. Fig. Error Message for Missing Image Upload.

27. Dept. of CSE, EWIT 27 3. Upload the photo. Fig. Image Uploaded Successfully for Damage Detection.

28. Dept. of CSE, EWIT 28 4. Side View Detection. 5. Front View Detection. Fig. Detected Damages and Cost Estimation for Side View. Fig. Detected Damages and Cost Estimation for Front View.

29. Dept. of CSE, EWIT 29 5. Rear View Detection. Fig. Detected Damages and Cost Estimation for Rear View .

30. Dept. of CSE, EWIT 30 6. Non-Damaged Car. Fig. No Damage Detected for Non-Damaged Vehicle.

31. CONCLUSION This project implements an automated system for detecting car damages and estimating repair costs using a web-based platform. It integrates Flask for managing user interactions, image uploads, and result display. The YOLO object detection model efficiently identifies damaged car parts from uploaded images, while a predefined price structure calculates repair costs. The system is user-friendly, providing real-time cost estimation with visualized damage results. Its modular architecture allows easy customization and scalability for future enhancements like database integration and expanded part detection. Dept. of CSE, EWIT 31

32. Dept. of CSE EWIT 32 FUTURE ENHANCEMENTS Damage Severity Classification – Identify minor, moderate, and severe damages. Multi-Angle Image Processing – Support for multiple images from different angles. Cloud-Based Deployment – Deploy the system online for public access. Real-Time Detection – Support video-based real-time damage assessment.

33. S. K. Aithal , K. C. Nackathaya , D. Poojary, G. Bhandary , and A. Acharya, "Automated vehicle damage detection and repair cost estimation using deep learning," in Proc. 2nd Int. Conf. Sustainable Comput . Smart Syst. (ICSCSS), pp. 1480–1484, 2024. doi : 10.1109/ICSCSS60660.2024.10625107. Y. Ma, H. Ghanbari , T. Huang, J. Irvin, O. Brady, S. Zalouk , H. Sheng, A. Ng, R. Rajagopal, and M. Narsude , "A system for automated vehicle damage localization and severity estimation using deep learning," in Proc. IEEE Int. Conf. Intell . Transp. Syst. (TITS), pp. 5627–5639, 2024. doi : 10.1109/TITS.2023.3334616. D. Mallios , L. Xiaofei , N. McLaughlin, J. M. D. Rincon, C. Galbraith, and R. Garland, "Vehicle damage severity estimation for insurance operations using in-the-wild mobile images," IEEE Access, vol. 11, pp. 78644–78655, 2023. doi : 10.1109/ACCESS.2023.3299223. Q. Zhang, X. Chang, and S. Bian , "Vehicle-damage-detection segmentation algorithm based on improved Mask RCNN," IEEE Access, vol. 8, pp. 6997–7004, 2020. doi : 10.1109/ACCESS.2020.2964055. U. Waqas, D. Lee, N. Akram , J. Jeon, and S. Kim, "Vehicle damage classification and fraudulent image detection including Moiré effect using deep learning," in Proc. IEEE Canadian Conf. Electr . Comput . Eng. (CCECE), pp. 1–6, 2020. doi : 10.1109/CCECE.2020.123456. REFERENCES

34. Dept. of CSE EWIT 34 G. Jocher , A. Stoken , J. Borovec , et al., " Ultralytics YOLOv5: State-of-the-Art Object Detection at Scale," in Proc. IEEE/CVF Conf. on Computer Vision and Pattern Recognition Workshops (CVPRW), pp. 1576–1584, 2022. doi : 10.1109/CVPRW56347.2022.00178. A. Kumar, S. Gupta, R. Singh, and P. Tiwari, "Deep learning-based vehicle damage detection and classification using transfer learning," Journal of Intelligent Transportation Systems, vol. 27, no. 3, pp. 345–358, 2023. doi : 10.1080/15472450.2022.2156789. L. Chen, Y. Wang, Z. Li, and H. Zhang, "A multi-task learning framework for vehicle damage detection and severity assessment," IEEE Transactions on Intelligent Vehicles, vol. 8, no. 2, pp. 1234–1245, 2023. doi : 10.1109/TIV.2022.3219876. M. Patel, R. Sharma, and K. Verma, "Real-time vehicle damage detection using YOLOv5 and IoT-enabled edge devices," in Proc. IEEE Int. Conf. on Artificial Intelligence and Robotics (ICAIR), pp. 456–461, 2023. doi : 10.1109/ICAIR.2023.10234567. T. Nguyen, H. Tran, and V. Le, "Enhancing vehicle damage detection with synthetic data generation and domain adaptation," in Proc. IEEE Conf. on Computer Vision and Pattern Recognition Workshops (CVPRW), pp. 789–795, 2023. doi : 10.1109/CVPRW.2023.4567890. J. Park, S. Lee, and H. Kim, "A comparative study of deep learning models for vehicle damage detection and cost estimation," Sensors, vol. 23, no. 5, pp. 1–18, 2023. doi : 10.3390/s23052567.

35. THANK YOU Dept. of CSE EWIT 35