Cloud—based multiple importance sampling algorithm with AI based CNN classifier for secure infrastructure

• Published in: Automated software engineering Vol. 28; no. 2; p. 16
• Main Authors: Dhaya, R., Kanthavel, R.
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.11.2021; Springer Nature B.V
• Subjects: Algorithms; Artificial Intelligence; Artificial neural networks; Body parts; Classification; Classifiers; Cloud computing; Computer Science; Human motion; Importance sampling; Infrastructure; Neural networks; Occlusion; Software Engineering/Programming and Operating Systems; Special Issue on Intelligent Software-Driven Autonomous Systems; Target recognition; Tracking; Two dimensional bodies; Occlusion; 2D image; 3D image; AI-CNN; Human upper body parts; MMIS; Dynamic environment
• ISSN: 0928-8910; 1573-7535
• DOI: 10.1007/s10515-021-00293-y

Enhancing Security infrastructure in the cloud data center has been a typical task that includes tracking human motion and their body parts under dynamic environment has been a difficult task during video detection situations. The research proposal's goal is to propose the human object video tracking methods to estimate the movement of human upper body actions and classify it under a dynamic environment. Moreover, the movement of upper body parts estimation includes face and arm variation or detections to define the classification problem. Considering the combination of frequent moving object parts with occlusion problems, decision making of target object identification and upper body pose variations will result in improving the classification accuracy. It also helps with a reduction in occlusion of the moving body parts detection. The modified multiple importance sampling filter with AI-based convolution neural network classifier has been proposed to track human poses with fast-moving actions. Dynamic sampling filer tracks the upper part of the human body with 2D images and 3D postures. Finally, similar poses are classified using an updated convolution neural network classifier which is designed for human object classification. The high accuracy of the system has been obtained for cluttered environments with occlusion problems by properly obtaining the sampling states of the filter as shown in the experimental result analysis part.