Research On Vehicle Intelligent Damage Location Algorithm Based On Deep Learning

In the traditional process of claim,the loss assessment is the most crucial step.Indeed,estimating the loss of vehicle by pictures is an essential method for the insurance company to conduct an assessment.However,there is some error in practical work.Human beings perform this work,and it relays on the experiences and skills of professional staff.Besides,professional staff needs to spend much time recording the loss assessment,which will not only make the process of assessment last long period but make a negative impact on the field to some extent.This dissertation adopts a method named intelligence vehicle.This method was realized by deep learning.First of all,recognizing and accessing visual damage parts in the pictures of the accident vehicle,especially,the subject needs to reorganize and improve the data,which makes sure the subject conducted efficiently.Besides,recognizing the structure of vehicle by MASK R-CNN Semantic segmentation and marking the specific broken part by Yolo-v3.According to combining recognizing the general structure and marking specific broken part,the system examines the broken part located in which part of the whole vehicle.Then,the marking and locating for the broken part in the process of loss assessment finished.The result of experiment shows that the Mask R-CNN and YOLO-V3 can put out the result of assessment efficiently and with high accuracy.In summary,this dissertation applies the deep-learning(image classification,object localization,semantic segmentation and instance segmentation)to the intelligence vehicle loss assessment.And it can be conducted in high accuracy,and short period,besides,a standardized assessment can be realized beyond the human-beings subjective judgement.