Enhancing Vehicle Inspection: A Comparative Study of VGG16, VGG19, Densenet and Mobile Net for Real-Time Car Damage Detection

Abstract

Due to the rising number of car incidents, the insurance industry is being overstretched right now by the fact that it has a lot of claims to attend to, and this poses claim leakage as a fast-growing occurrence. On the other hand, AI-driven processes that use machine learning and deep learning techniques have turned out highly effective in solving these challenges. The subject of this research is exactly how to redesign existing damage assessment systems and making them better with the help of machine learning models. In this paper, an evaluation of four deep learning-based algorithms' performances in classifying vehicle damage, consisting of VGG16, VGG19, DenseNet121 and MobileNet. Each of these models are trained independently and the assessment of their accuracy is carried out to decide the most effective one. To enhance user accessibility, the adoption of an interface running under Streamlit where users have an opportunity to upload any image for assessment of the sustainability of the vehicle. This interface will offer accurate predictions of existing car damages together with probability values of their occurrence. In addition, the study has implemented a camera live feed feature with the main interface of the system such that users can observe damage detection by turning their smartphone camera on. This novel method reduces the possibility of the claim reprocessing cycle and significantly reduces the possibility of misreports by determining the damaged vehicles and non-vehicles accurately.