Location Detection of Informative Bright Region in Tunnel Scenes Using Lighting and Traffic Lane Cues

Abstract

To locate Informative Bright Region (IBR) in which visual information is missing owing to limited dynamic range of image sensor, an algorithm is proposed that combines the geometric properties of visual cues into a confidence map. The location of an IBR in a road tunnel scene is estimated in real-time under the condition in which most of the vision information inside the IBR is lost. The algorithm is evaluated by comparing the estimated location of IBR with that annotated by multiple human observers in a self-built tunnel scene video dataset recorded by a car-mounted camera, and the algorithm achieves a running time of 10 ms for each frame. The algorithm aims to provide control timing of imaging sensor on a low-cost platform such as a vehicle driving recorder to enhance the visual contents captured in over-exposed regions.