Abstract
The brain is a typical complex system that executes visual information analysis, motor control, selective allocation of memories, and so on. We have developed a software platform to simulate the complex brain system numerically by computational models, especially focused on vision. Our platform for vision simulation bases on RT-middleware and OpenRTM-aist, which is a software platform to develop robotic system. A new datatype as a common interface of various vision models is provided. The new datatype and our software library enable automatic switching of transformation method of input/output data between vision models, i.e., shared memory or via computer network. We also provide a software package named by OpenCV-RTC which converts a lot of image processing functions of OpenCV into RT-components executable on OpenRTM-aist. We show that a novel model for estimation of fixation location of humans' eye is efficiently developed on our platform by parallel connection of existing models for eye fixation, and show that the new model significantly outperforms the existing models.
