3D scene reconstruction and object recognition for use with autonomously guided vehicles (AGVs)

Sensor modelling of robot and autonomously guided vehicles (AGVs) workspace is an important requirement if semi-intelligent or intelligent machines are to be maximized to their full potential whilst operating in unconstrained environments. A number of advances have led to the development of sub-systems which are used in part or wholly to aid environment information. This paper outlines a novel vision system intended for use on AGVs to provide object recognition and collision avoidance through the use of discrete image processing and abstract feature representation. The working prototype contains a single CCD camera and cooperating manipulator to provide accurate 3D positioning. Depth information is determined by the use of inverse perspective projection using high level features obtained during acquisition. In this way solid objects may be modelled as wire frames and used during object recognition. The wire frame representation of an object allows simple geometric shape fitting to be used. Objects are stored as a collection of 2D geometric shapes which is then used to potentially identify an object in the field of view. Working examples from the prototype are used to aid description.