A unified model of video-based human action categorization using Chaotic Quantum Swarm Intelligence on Intuitionistic fuzzy 3D Convolution Neural Network

摘要

In the contemporary surveillance schemes of Computer Vision, videos concerning human action categorization have become a predominant zone, involving Pattern Recognition tasks. Factually, most of the human actions comprise complex temporal information, and it is quite difficult to discover the diverse activities of humans precisely, in an unpredictable variety of environmental circumstances. A Deep Learning paradigm can tackle this issue, by providing additional capabilities to vision-based human action recognition. However, there are more complex challenges in extracting the spatio-temporal features, for instance, the presence of noise in videos and the highly vague feature points. This paper proposes a hybrid intelligent Intuitionistic Fuzzy 3D Convolution Neural Network that uses Chaotic Quantum Swarm Intelligence (CQSI-IFCNN), to optimize video-based human action categorization. Vagueness and ambiguity of input video frames are inherited by Intuitionistic Fuzzy networks in terms of membership, hesitation and non-membership components. By applying Chaotic Quantum Swarm Intelligence (CQSI), the learning parameters and error rates that occur in standard convolutional neural network are considerably reduced. The chaotic searching scheme is applied to overcome premature local optima in Quantum Swarm Intelligence. Therefore, this model produces optimized outcomes in Intuitionistic fuzzy 3D Convolutional Neural Networks, thus improving the categorization of human actions in videos. The Performance of CQSI-IFCNN is assessed by using the KTH and UCF Sports Action datasets. From the simulation outcomes, it is observed that CQSI-IFCNN has attained a higher rate of action categorization accuracy than standard CNN and PSO-CNN.