Abstract:
The potential to move in surrounding and execute designated work by the autonomous
systems necessitates accurate position estimation and orientation of mobile node.
Advancement in the field of robotics and computing systems needs incorporation of fast and
accurate location information. Current localization techniques are complex, computationally
heavy, expensive, less accurate and requires more power, processing speed. Therefore,
formulation of cheap and efficient localization scheme for Robots, which should also be
capable to replace GPS and provide better/ comparable results relative to other localization
techniques, is the objective of proposed work.Various localization techniques including radio signal (Wi-Fi received signal strength wirelesslocal area network based system, Global positioning system), Odometry information andvision based systems are analyzed and a vision based localization scheme is proposed. Theproposed technique is based on the shape/size changes of single object with its perspectiveview. It incorporates projective geometry concepts and elaborates that how shape/ sizechanges of an object by varying distances/ view angle withchanged position provides distinctposition coordinates. It also spotlights an idea that the shape of object with respect to itssurrounding is invariant to the changes of the view point and distances and it is also invariantto the camera intrinsic parameters (as the camera is uncalibrated). The problem has alsobeen modeled to confirm authenticity of proposed solution.
The testing of proposed technique has been carried out at 10x12 feet indoor, outdoor
environments and an error upto 4 pixels= 4 x 0.3175 = 1.2698cm (as one pixel= 0.3175cm
approx) has been obtained. Proposed technique has also been compared with existing
techniques and their various features have been critically brought to lime light.
