Research on Performance Evaluation Methods for Event Detection Technology based on Onboard Camera Images
Ministry of Land, Infrastructure, Transport and Tourism, Japan
In recent years, private sector has significant progress in the development of dashcam technology etc.
This project, making use of onboard camera images from dashcams,
researched on performance evaluation of the event detection technology towards image creation and the evaluation method,
with a focus on detecting objects falling from vehicles and potholes on roads etc.
To be more specific, images are captured when events like falling objects or potholes are detected during the patrol of highway maintenance vehicles. Based on these images collected in the real environment, we produced images for performance evaluation. Besides, methods of evaluating detection performance of target events, which are essential for the consideration of implementing an event detection system using onboard camera images, is reviewed. A performance evaluation methodology was prepared after collecting opinions from the collaborators who process image detection technology and verifying the performance evaluation methods.
Devices installation environment:
A scene of installing devices in the vehicle for experiment
Driving pattern which object detected is on adjacent lane
Driving pattern which object detected is on own lane
An image collected from a vehicle on patrol
(circle in red dedicates the event detected)