Augmentation and Detection of Individual Pose using CUDA. Issue 1 (February 2021)

• Record Type: Journal Article
• Authors: Muzammil Parvez, M; Inthiyaz, Syed; Praghash, K; Suresh Kumar, K; Sreevardhan, cheerla
• Is Part Of: Journal of physics. Volume 1804:Issue 1(2021)
• Publication Date: 2021-02
• Subjects: Physics -- Congresses; 530.5
• DOI: 10.1088/1742-6596/1804/1/012178
• Languages: English
• ISSNs: 1742-6588

Abstract

Abstract: The collective real-time pose prediction is a leading element in allowing algorithms to understand Individuals in videos and pictures. In this study, we are presenting a strategy to determining the pose of various individuals in a picture in real time. Non-parametric procedure is used as representation that we refer this to Part-Affinity-Fields (PAFs) to understand how to connect parts of the body with individuals. This base up framework accomplishes high exactness and real time execution, paying little mind to the quantity of individuals in the picture. In past work of computer vision researchers, PAFs and body part area estimation were refined at the same time across preparing different steps. We show that PAF just clarifying as opposed to both the PAF and content part area refinement brings about a significant increment in each of runtime execution and precision. We likewise propose the main joined body and foot key point identifier, considering an inner commented on foot dataset that we have freely discharged. The joined finder not just diminishes the derivation time contrasted with running them successively, yet in addition keeps up the exactness of every part separately. The work was completed in appearance to Open Pose, chief opensource continuous structure to multiindividual 2D present disclosure, including foot, body, hand, & facial central issues.