Computer Vision based Automated Spare Part Finder App

Abstract

The “Automated car spare part finder app” aims at building a spare part detection mobile application using image recognition technique that will assist users in finding desired spare parts and provide a list of websites where those spare parts can be found. The developed system accomplishes the goal of integrating a deep learning model with a mobile application. A YOLOv5s model was trained on a custom dataset (spanning over 8 classes), which was built from scratch due to the unavailability of a spare part dataset sample. Following the training recommendations of YOLOv5, two models were generated by varying the run parameters and the data augmentation techniques, whilst maintaining the default model architecture. A lite version of the trained model was deployed on a mobile application, implemented using Flutter. The application allows users to select an image and run predictions on the selected image to identify and classify the part if present. Following detection, the application prompted users to input the make and model of the vehicle which would then be coupled with the predicted part name to search the web and generate a list of spare part suppliers for the specified part. This was achieved by using a web scrapper also implemented in Flutter. The resulting system, has an accuracy percentage of around 80% and has accomplished all requirements set.