Human gesture recognition based on image sequences

Human gesture recognition based on image sequences is now the research focus. In this study, four classifiers with K-NN, Bayes, LDA and SVM were used, two human gestures (walk and bend) were recognized. This study extracted the human body contour of image sequences, the distances between the contour and center of human body were used as input features. The results with a 5-fold cross-validation indicate that the classification accuracies of SVM and LDA are better than those of KNN and Bayes. This study also calculated the AUC values (the area under the ROC curve), the same results were obtained.