Modeling, Evaluation and Control of a Road Image Processing Chain [chapter]

Yves Lucas, Antonio Domingues, Driss Driouchi, Pierre Marché
2005 Lecture Notes in Computer Science  
Tuning a complete image processing chain (IPC) remains a tricky step. Until now researchers focused on the evaluation of single algorithms, based on a small number of test images and ad hoc tuning independent of input data. In this paper we explain how, by combining statistical modeling with design of experiments, numerical optimization and neural learning, it is possible to elaborate a powerful and adaptive IPC. To succeed, it is necessary to build a large image database, to describe input
more » ... es and finally to evaluate the IPC output. By testing this approach on an IPC dedicated to road obstacle detection, we demonstrate that this experimental methodology and software architecture ensure a steady efficiency. The reason is simple: the IPC is globally optimized, from a large number of real images (180 out of a sequence of 30 000) and with adaptive processing of input data Adaptive Processing in Vision Systems Designing an image processing application involves a sequence of low and medium level operators (filtering, edge detection and linking, corner detection, region growing ...) in order to extract relevant data for decision purpose ( pattern recognition, classification, inspection ...). At each step of the processing, tuning parameters have a significant influence on algorithm behavior and the ultimate quality of results. As the processing power of micro computers has reached a very high level, artificial vision systems are now developed for demanding applications such as video surveillance or car driving where scene contents is uncontrolled, versatile and rapidly changing. The automatic tuning of the IPC has to be solved there, as the quality of low level vision processes should be continuously preserved to guarantee high level task robustness. The first problem to solve in order to design adaptive vision systems is the evaluation of image processing tasks. Since a few years, researchers are interested in it and proposed rather empirical solutions [1, 2, 3, 4, 5, 6, 7] . When a ground truth is available, it is possible to compare directly this reference to the results using a specific metric. Sometimes
doi:10.1007/11499145_109 fatcat:kkrzdobc5rbohi6gl3dqbpyvfi