Abstract:
The CarSight project focuses on developing an intelligent mobile application designed to transform the used car market in Pakistan using machine learning and image recognition technologies. The primary objective is to provide accurate price assessments by analyzing vehicle characteristics and market trends, thereby eliminating the subjectivity and inconsistency prevalent in traditional pricing methods. CarSight integrates the YOLOv8 model for real-time identification of a vehicle’s make, model, variant, year, and condition, as well as for detecting odometer readings, using EasyOCR for text extraction. Additionally, the application utilizes XGBoost and LightGBM models for price estimation. Custom datasets were developed for vehicle identification, odometer region detection, and price estimation. The models were trained on over 12,000 vehicle images for identification, 1,250 images for odometer detection, 9,500 images for damage detection, and 21,000 vehicle price records for price estimation, all sourced from local markets and fine-tuned for accuracy. The YOLOv8 model demonstrated strong performance in vehicle and odometer region identification, achieving a precision of 90% and a mean Average Precision (mAP@0.5) of 94.9% for vehicle identification, as well as a precision of 94% and mAP@0.5 of 94.9% for odometer detection on test sets. YOLOv8 also played a key role in evaluating vehicle condition, achieving a precision of 93.4% and mAP@0.5 of 90.5%. For price estimation, the XGBoost and LightGBM models performed exceptionally well, with XGBoost achieving an R-squared value of 0.98 and the lowest mean absolute error (MAE) on the test set. These results confirm CarSight’s effectiveness in providing precise vehicle identification, odometer readings, and market price estimates. Furthermore, the user-friendly design of the CarSight mobile application ensures accessibility to a wide range of users, empowering both buyers and sellers to make informed decisions based on reliable data.
