Marek Samulka, Peter Kostelnik, Miroslav Hudec, Peter Sincak
Center for Intelligent Technologies
Computational Intelligence Group,
Faculty of Electrical Engineering and Informatics,
Technical University of Kosice, Slovakia
(Poster contribution)
Biological Inspiration for neural technology has various degree of implementation and its application to brainlike system is very challenging problem. There are number of different approaches to these neural technologies including ART-like neural networks, various Malsburgs models and the whole family of Neocognitrion systems based on Hubel and Wiesel model (H-W model) of the visual cortex and its information processing.
The presented work represents the experience with neocognitron technology developed by Prof. Fukushima in late 80-ties to identify the type of positions of the mobile robots on the soccer playground. This information should be complementary to one of the selected players team and should represent and intelligent advisor for the coach of the selected team to evaluate a type of situation on the playground. The input should be an image of the playground and the output should be a categorization to which type of situation is selected team heading to e.g. attacking formation, defending formation and so on. This information should be considered as additional for selected team.
Neocognitron is a multi-layer and hierarchical neural complex, which is modeling the visual system, and essentially provide tool for pattern recognition that is robust for different positions of the patterns. This neural network is considered as a biologically inspired system with the high degree of biological interpretation.
Basically the multiplayer neural network is based on self-organizing elements and generally the approach is based on various types of layers and cells as follows:
The network consists of “n” twin layers (S and C, n is usually 4) which are processing the input. The network is in fact calculation mathematical transformation, which extract features that are invariant to rotation and shifting of the object. Different positions of the robots are a good occasion to demonstrate the use of this approach. The input of the system is an image and is calculated in parallel and identification of different robots on the image are expected for further decision procedures. Input image is transformed into the image of linked curves, which were made by connections of the robots (players) of both teams respectively. The created 2 curves represent the position (formation) of robots in each team. This new image is very similar to characters or letters and is associated with the configuration of the players in the sense of attacking position, defending position or any strategic distribution of the players. So the output of the neural network is categorization the types of the players’ distribution on the field related to strategic positioning. This identification of the situation can be useful for decision support system related to further actions of the selected team. Experience with implementation of this system and application to multi-robot environment are presented. The results show the ability to use this rather complex and complicated technology for these purposes.
The biological inspiration is here very strong and presents a model of the visual system for real-time processing requirements. The main aim of this poster is to present experience with implementation and application of neurocognitron neural network in the problem of estimation of the type of the strategic distribution of soccer players on the field and its identification by the help of biologically inspired system. The similarity of this task with the letter-character recognition problem (which was successfully accomplished by neocognitron) is presented in the project.
Submission as poster contribution – topic : VISION, Neural Models, Other application