Multiple 3-D Objects Detection and Recognition Using Independent Component Analysis

In this paper we present a new approach to the detection and recognition of multiple 3-dimensional objects in an image and describe a working, near-real-time multiple 3-D objects detection and recognition system. The developed method processes in three steps. At the first step the edges of the objects in an image are found using edge detection following which the boundaries are detected using a block processing algorithm (BPA), traced and labelled. In the second step the 3-D object recognition of the detected objects is accomplished by utilising neural-network architecture developed based on independent component analysis (ICA). In the last and the crucial phase of the process, an algorithm is developed where any unknown object (the object that is not available in the database) is added to the existing database followed by dynamically training the ICA neural network, thereby making the database ever growing. The system developed is also capable of detecting and recognizing objects that are occluded .The database of images captured by a CCD camera is used in this paper.