Gesture based control of mobile robots

This paper presents guidelines to achieving robust gesture recognition by utilizing a classification method which constructs classifiers in a cascaded structure. This can be achieved by obtaining a robust cascade for hand detection from the training algorithm proposed. This has been implemented onto a gesture recognition system which was developed for the purpose of gesture based control of mobile robots. An experimental approach was adopted in identifying the dominant training parameters based on observations on the performance of the cascades. Four parameters were suspected to have significance on the training results. It is computationally expensive to evaluate all possible combinations of different levels of the parameters and to perform analysis on the data to justify if the hypothesis made is correct. The Taguchi Method, was used to carry out the analysis. The main insight here is that investigation was carried out without having to carry out the entire set of possible combinations, analysis was performed by carrying out a fraction of the entire set of possible combinations according to a special pattern which is specified by Taguchi Method. Analysis of Variance (ANOVA) was used to analyze the training data. Results of F Test reveal the correctness of the hypothesis made earlier. Further works can be carried out based on results of this paper to obtain optimal values for the dominant parameters which have been identified and the optimal values can be used as input to the training algorithm to obtain a robust cascade for hand detection.