Vergence control for a biologically inspired binocular active vision system

• Main Author: Zhang, Xuejie
• Other Authors: Tay Leng Phuan, Alex
• Format: Theses and Dissertations
• Language: English
• Published: 2012
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence; DRNTU::Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision
• Online Access: https://hdl.handle.net/10356/50632
• Institution: Nanyang Technological University

The human naturally possesses a robust and effective binocular vision system that utilizes saccade and vergence eye movements to explore the visual environment. This thesis studies how low-level visual functions such as vergence and saccade can provide a basis to build an active vision system that is able to robustly support scene exploration. Several available vision techniques were examined for suitability of application into human-like binocular active vision systems. A biologically inspired binocular active vision proof-of-concept platform was designed and built. This thesis focuses on the development of biologically inspired vergence control models. Vergence is the capability of human vision to gaze two eyes on a single spot in space. Through the study on the human vision system, several vergence control models were developed for the binocular vision platform. An initial vergence control model based on edge features in log polar space was first developed to achieve vergence in a simplified environment. Through further research and experimentation, investigations progressed onto a pyramidal vergence control model using disparity energy neurons. This second model utilizes fixed tuning disparity energy neurons and a pyramidal image structure to estimate disparity, resulting in near real time vergence control. Subsequently a pyramidal area correlation model was developed for simple vergence capabilities. As the search for a better solution prevailed, the model evolved into a disparity estimation model using a pyramidal fusion of log polar images. This final model superseded the previous models and exhibited robustness, providing a fundamental basis for developing real world applications for binocular foveation without object registration techniques or very accurate camera lens calibrations. Besides the developed vergence control models, the breadth of supporting technologies included the building of mechanisms that simulated saccadic eye movements through the use of image segmentation and visual attention approaches. Through the study and development of the biologically inspired components, the developed binocular vision system was able to select fixation points, reconstruct depth of the attended positions through vergence and build a simple fixation based spatial understanding of the environment. This thesis is presented with a balance of research rigour in the vergence domain and a broad-based understanding of human visual concepts dealing with viable computational solutions to saccade and vergence.