Fuzzy Integral and Cuckoo Search Based Classifier Fusion for Human Action Recognition

The human activity recognition is an important issue for sports analysis and health monitoring. The early recognition of human actions is used in areas such as detection of criminal activities, fall detection, and action recognition in rehabilitation centers. Especially, the detection of the falls in elderly people is very important for rapid intervention. Mobile phones can be used for action recognition with their built-in accelerometer sensor. In this study, a new combined method based on fuzzy integral and cuckoo search is proposed for classifying human actions. The signals are acquired from three axes of acceleration sensor of a mobile phone and the features are extracted by applying signal processing methods. Our approach utilizes from linear discriminant analysis (LDA), support vector machines (SVM), and neural networks (NN) techniques and aggregates their outputs by using fuzzy integral. The cuckoo search method adjusts the parameters for assignment of optimal confidence levels of the classifiers. The experimental results show that our model provides better performance than the individual classifiers. In addition, appropriate selection of the confidence levels improves the performance of the combined classifiers.