Emergence of deep learning has significantly improved the performances of machine learning models, and various pre-trained models have made life easy for organizations that intended to accomplish their task with the help of deep learning-based systems. The aim of this research is to conduct a comparative study of a few popular pre-trained deep learning models based on a primary dataset. The key contribution of this work lies in creating the dataset and conducting the comparative analysis of the pre-trained models. Raw data were initially collected from various accident spots and from workshops where people usually go to repair their damaged vehicles. Later, these datasets were validated by a popular insurance company of the country. VGG16, CGG19, ResNet50, InceptionV3, and MobileNet are the pre-trained models used in this study for their well-accepted performances in the computer vision-based problems. After initial preprocessing which includes removing bad images, resizing, and rescaling, the dataset was prepared for the training. Eighty percent of the dataset was used for training the model and 10% each for testing and validation. After testing the models, a few fine-tuning was done to enhance the performances of the models. Final observation reveals that, out of the five models, InceptionV3 outperformed all other models with an accuracy of 97%. Finally, the performance of the proposed model was compared with other recently published models and the proposed model has outperformed those also. Outcome of the research can provide a more dependable and efficient method for handling automobile insurance claims after accidents.