부산대학교 도서관

In today’s world, with all the technological advancements that we have made, the process of car damage detection is a very lengthy process involving a lot manual work. It requires human verification approach which is very slow and leads to an extremely arduous and tardy process, which leads to human error creeping into the results, thereby increasing the hardships of the common man. Proliferation of Indian automobile industry is directly proportional to the car incidents which is also directly proportional to more insurance claims. Insurance companies need to cover many simultaneous claims and solve the issue related to claims leakage. We explore different DNN based techniques for the purpose of the vehicle damage detection which will completely eliminate the large amount of paper work and man power for physical damage estimation and shift this entire process to an efficient AI based solution which can provide a rapid claim process in a shorter time span. The proposed model provides high accuracy confidence scores for the detected damages which are classified on the basis of the 21 vehicle damage classes that we have defined so that there can be an extensive segregation of damages incurred by the vehicle.