Open Access Thesis
Date of Award
Electrical and Computer Engineering
More and more drivers nowadays enjoy the convenience brought by advanced driver assistances system (ADAS) including collision detection, lane keeping and ACC. However, many assistant functions are still constrained by weather and terrain. In the way towards automated driving, the need of an automatic condition detector is inevitable, since many solutions only work for certain conditions. When it comes to camera, which is most commonly used tool in lane detection, obstacle detection, visibility estimation is one of such important parameters we need to analyze.
Although many papers have proposed their own ways to estimate visibility range, there is little research on the question of how to estimate the confidence of an image. In this thesis, we introduce a new way to detect visual distance based on a monocular camera, and thereby we calculate the overall image confidence.
Much progresses has been achieved in the past ten years from restoration of foggy images, real-time fog detection to weather classification. However, each method has its own drawbacks, ranging from complexity, cost, and inaccuracy.
According to these considerations, the new way we proposed to estimate visibility range is based on a single vision system. In addition, this method can maintain a relatively robust estimation and produce a more accurate result.
Huang, Minglei, "Visibility And Confidence Estimation Of An Onboard-Camera Image For An Intelligent Vehicle" (2015). Wayne State University Theses. 425.