An AI model detects areas of exterior vehicle damage, such as scratches and dents, in captured photos. [Photo: Humax Mobility]

Humax Mobility affiliate Kyle E3 Jasper is moving to verify vehicle damage reading technology. Kyle E3 Jasper is working with a Seoul National University laboratory led by Professor Lee Jang-woo (이장우) and AI company Hodu AI Lab to conduct a technology pilot of an "exterior vehicle defect detection system," it said on Wednesday.

The system analyses photos taken when car-sharing customers pick up and return vehicles to determine whether exterior damage exists and where it is. The aim is to secure in advance vehicle operation technology needed in the robotaxi era. The company said return inspections for car-sharing vehicles are still handled by people. Because inspectors check renters' photos and records one by one, the time and staffing burden increases as the number of vehicles in operation rises and the number of photos to review grows.

The three organisations will develop a system in which AI analyses photos and videos taken from multiple angles to first flag suspected accident vehicles and damaged areas. They plan to develop technologies including object detection that marks suspected damage locations in photos, segmentation that shows the shape and area of scratches and damage at the pixel level, and image registration that aligns past photos or normal-vehicle images to the same position and size as current photos to find changes. The company expects inspection efficiency and reliability to improve.

The pilot will be carried out in two stages. In the first stage, it will develop a prototype and collect data, train detection and segmentation models and build an inference pipeline. In the second stage, it will expand the system and improve performance, while studying the feasibility of analysing vehicle changes using image registration.

In preliminary validation, the exterior analysis model posted about 85 percent accuracy in distinguishing normal from defective images on evaluation data comprising 100 normal photos and 100 defective photos. In this pilot, it plans to build data aligned with actual inspection standards and advance the model to raise the feasibility of on-site application.

Kyle E3 Jasper will review ways to apply the results in real operating environments based on the pilot. It also plans to examine possible linkages with group affiliates, including the Humax Mobility Group's "robotaxi fleet operations infrastructure joint business consultative body" launched last month.

Song Yoon-hwa (송윤화), CEO of Kyle E3 Jasper, said the pilot is a process to secure in advance "automatic vehicle condition reading" technology needed for fleet operations in the robotaxi era by adding AI technology to know-how accumulated over the past 28 years in vehicle operations and maintenance and management. "Through this technology, we will enable stable management of vehicle conditions even in environments operating robotaxis or autonomous driving fleets," Song said.

Keyword

#Humax Mobility #Kyle E3 Jasper #Seoul National University #Hodu AI Lab #robotaxi
Copyright © DigitalToday. All rights reserved. Unauthorized reproduction and redistribution are prohibited.