Using the conflict in Dempster-Shafer evidence theory as a rejection criterion in classifier output combination for 3D human action recognition

In this paper, we propose a comprehensive solution to 3D human action recognition including feature extraction, classification, and multiple classifier combination. We effectively present two feature extraction methods, four different types of well-known classifiers, and four multiple classifier combination strategies including a specially designed belief based method. In order to enhance the recognition accuracy, we propose a new rejection criterion based on the conflict from the information sources: the classifier outputs. We test our method on the MSRAction 3D dataset. Discarding examples using the conflict based criterion shows superior results than other combination approaches. Moreover this criterion allows choosing a tradeoff between the performance and rejection rate.