Improving active vision system categorization capability through Histogram of Oriented Gradients

In the previous work of Mirolli et al. [1], an active vision system controlled by a genetic algorithm evolved neural network was used in simple letter categorization system, using gray-scale average noise filtering of an artificial eye retina. Lanihun et al. [2] further extends on this work by using Uniform Local Binary Patterns (ULBP) [4] as a pre-processing technique, in order to enhance the robustness of the system in categorizing objects in more complex images taken from the camera of a Humanoid (iCub) robot . In this paper we extend on the work in [2], using Histogram of Oriented Gradients (HOG) [5] to improve the performance of this system for the same iCub image problem. We demonstrate this ability by performing comparative experiments among the three methods. Preliminary results show that the proposed HOG method performed better than the ULBP and the gray-scale averaging [1] methods. The approach of better pre-processing with HOG gives a representation that could translate to improve motor responses in enhancing categorization capability for robotic vision control systems.

This is an Accepted Manuscript version of the following work: Lanihun, O., Tiddeman, B., Tuci, E., Shaw, P. (2015). Improving Active Vision System Categorization Capability Through Histogram of Oriented Gradients. In: Dixon, C., Tuyls, K. (eds) Towards Autonomous Robotic Systems. TAROS 2015. Lecture Notes in Computer Science(), vol 9287. Springer, Cham.. This version of the manuscript has been accepted for publication, after final editorial and peer review (where applicable) is complete, but is not the Version of Record and does not reflect post-acceptance improvements (such as copyediting or typesetting), or any corrections. The final authenticated version is available online at: http://dx.doi.org/10.1007/978-3-319-22416-9_16 . Use of this Accepted Manuscript version is subject to the publisher’s Accepted Manuscript terms of use: https://www.springernature.com/gp/open-research/policies/accepted-ma...