July 22, 2019
When vehicles are handed over from one owner to another, manual quality inspection is mostly required. With the growing amount of vehicle rental companies and shared car services, the need for automatic vehicle inspection grows. We have developed a deep learning model, able to detect and locate 13 different damage types on a vehicle. To do so, we used thousands of images containing damaged vehicles, and learned the computer to detect the different damage types. We proved that the model is as accurate as humans when it comes to locating and classifying the damage.
We developed this technique together with Pon Logistics to improve their inbound vehicle damage detection. As they are processing roughly 120,000 new vehicles a year, scalability and accuracy are of importance. Using this system, Pon Logistics is able to more accurately detect vehicle damages and to improve the process flow. The major advantage is that the damage detection can be performed within seconds, removing unnecessary waiting time for customers.
Using this technique, we are able to give a consistent damage report for each vehicle. As it is designed to work on low-quality images, it is easily implemented at different locations using existing cameras or by integrating it in applications on the mobile phone.
Robin van RuitenbeekAI Specialist