CHAN YAN SHENG
Aznul Qalid Md Sabri
Fylix Sdn Bhd
Automated vehicle inspection is a technique which requires technology from computer vision and deep learning to comprehensive and accurate inspections of vehicles. Compared to manual inspection, this technique can help to reduce the time and cost of inspection as well as increase the inspection accuracy and efficiency. In this project, an automated vehicle inspection system is developed using convolutional neural network (CNN) and explainable AI (XAI) to detect different types of damage of vehicle such as dent, scratch, crack, glass shatter, lamp broken and tire flat, and label the damages to respective car part. There are total of 3 models used in this project and the model used in this project is YOLO for damage detection model and Local Interpretable Model-agnostic Explanations (LIME) as XAI model to explain and verify the result of the damage detection to convince the user of the developed inspection system. To do the vehicle inspection, user will have to upload an inspection video and the model will analyse the video to get the inspection result and export the inspection report to user. The accuracy of the trained model overperformed other state-of-the-art method such as DCN and HTC which use ResNet-101 as the backbone of the model.