Automated Parameter Tuning in Stereo Vision Under Time Constraints

This paper presents a method for tuning parameters under a fixed time constraint for a general binocular stereo-vision algorithm. A major difficulty in stereo vision, as well as in other vision algorithms, lies in adjusting the large variety of parameters for maximizing performance. This effort is usually performed by human experts with a minimum of formal guidelines. To automate this process, we develop TEACHER 4.2, a generate-and-test system that systematically generates new parameter values by analyzing the results of previous tests, and that performs limited and controlled tests on the candidates generated using high-speed computers. The system is modeled as a statistical selection problem operating under a given time constraint. It divides the time allowed into stages, where promising parameter-value sets found in one stage are passed to the next stage for further testing, and selects the parameter-value set deemed best by the final stage as the result. We show experimentally tha...

• Type: Conference
• Year: 1992
• Where: ICTAI
• Authors: Steven R. Schwartz, Benjamin W. Wah
• Keywords: Artificial Intelligence, ICTAI 1992, Parameter-value Sets, Parameters, Time Constraint