Recognition of Overlapping Targets Using Artificial Neural Networks

Abstract

Image processing has been used in robot vision as a tool to establish the position of targets. Many algorithms using the template matching and artificial neural networks have been proposed for recognition. However, these algorithms are ineffective in computing cost and recognition accuracy in the case of recognition of overlapping targets. Therefore, we propose an effective algorithm for the recognition of overlapping targets using artificial neural networks which have learned movements of the human eye and the template matching. The artificial neural networks extract target candidates faster than conventional algorithms which find the candidates by moving templates of basic patterns in order. Then the template matching recognizes whether the candidates are targets or not. We compared the proposed algorithm with the conventional template matching algorithm with respect to the performance time and the recognition accuracy in the recognition of overlapping plant cells. The experimental results showed that the proposed algorithm recognized them faster than conventional algorithms and with high recognition accuracy.