A Neural Network Based on Biological Vision Learning and Its Application on Robot
This paper proposes a neural network called “Hierarchical Overlapping Sensory Mapping (HOSM)”, motivated by the structure of receptive fields in biological vision. To extract the features from these receptive fields, a method called Candid covariance-free Incremental Principal Component Analysis (CCIPCA) is used to automatically develop a set of orthogonal filters. An application of HOSM on a robot with eyes shows that the HOSM algorithm can pay attention to different targets and get its cognition for different environments in real time.
KeywordsNeural Network Receptive Field Input Image Training Phase Video Frame
Unable to display preview. Download preview PDF.
- 1.Hubel, D., Wiesel, T.: Receptive Fields, Binocular Interaction and Functional Architecture in the Cat. s Visual Cortex. J. of Physiology 160, 106–154 (1962)Google Scholar
- 2.Shou, T.: Brian Mechanisms of Visual Information Processing. Shanghai Science and Education Publishing House (1977)Google Scholar
- 3.Zhang, N., Weng, J., Zhang, Z.: A Developing Sensory Mapping for Robots. In: Development and Learning, 2002. Proceedings of the 2nd International Conference, June 12-15, pp. 13–20 (2002)Google Scholar