Building object models through interactive perception and foveated vision

Abstract

Autonomous robots that operate in unstructured environments must be able to seamlessly expand their knowledge base. To detect and manipulate previously unknown objects, a robot should be able to acquire new object knowledge even when no prior information about the objects or the environment is available. Additional information that is needed to identify new objects can come through motion cues induced by interactive manipulation. In the proposed system, changes in the scene are caused by a teacher manipulating the object to be learned. We propose to improve visual object learning and recognition by exploiting the advantages of foveated vision. The proposed approach first creates object hypotheses in peripheral stereo cameras. By directing its attention towards the identified object area in the foveal views, the robot can conduct a more thorough investigation of a smaller area of the scene, which is seen in higher resolution. We compare two methods for validating the hypotheses in the foveal views and experimentally show the advantages of foveated vision compared to stereo vision with a fixed field of view.

Compound eye and retina-like combination sensor with a large field of view based on a space-variant curved micro lens array

“…Similarly, the third vertex (labeling III) is transferred to the coordinate (16,10). Because the 4th vertex (labeling IV) of the pentacle is located in the ellipse area of both the 21th sector and the 22th sector in the 10th ring, it is transferred to the coordinates (21,10) and (22,10). Similarly, the 5th vertex (labeling V ) of the pentacle is transferred to the coordinates (28,10) and (29,10).…”

“…One of its remarkable merits is a retina-like feature that has space-variant resolution resulting from a nonuniform distribution of the photoreceptor cells in the retina. Because of the retina-like structure, the relationship between the retina and visual cortex submits to an approximate logarithmic-polar transformation (LPT) [18,19], which is beneficial for reducing redundant compression and improving the image efficiency of object detection and recognition [20][21][22][23]. To the best of our knowledge, there have been quite a few previous studies on both the compound eye and the single aperture eye [11,22].…”

Modeling and simulations on retina-like sensors based on curved surface

“…The advantages of retina-like sensors, including multi-resolution sampling [5], log-polar transformation (LPT) from retina to visual cortex [6,7], and invariance of scaling and rotation [8], etc., are beneficial to significantly compress redundant information and track objects in the large field of view (FOV) with high speeds [9,10]. Therefore, the techniques based on retina-like features are widely used in situations which require high resolution, large FOV, and real-time data processing at the same time, such as robotic vision [11], smart monitoring [12], vehicle navigation [13], etc.…”