Visual Memory Construction for Autonomous Humanoid Robot Navigation

A visual memory (VM) is a topological map that represents an environment as a direct graph of key images. Thus, visual information acquired from cameras onboard the robot are the only data to construct the map. This work presents the construction of a VM suited for the humanoid robot navigation framework. Additionally, a genetic algorithm that estimates the epipolar geometry is proposed to tackle the problem of image matching used within the VM construction process. Experimental results using a humanoid robot dataset are presented to validate the efficacy of our approach. Further, the solution for image matching based on the proposed genetic algorithm was compared with RANSAC.