کنفرانس بین المللی فناوری اطلاعات و دانش

English

صفحه اصلی / دوازدهمین کنفرانس بین المللی فناوری اطلاعات و دانش

Designing an AI-assisted toolbox for fitness activity recognition based on deep CNN

نویسندگان :

Ali Bidaran¹ Saeed Sharifian²

1- دانشگاه صنعتی امیرکبیر 2- دانشگاه صنعتی امیرکبیر

کلمات کلیدی :

Fitness AI-assisted, Magnitude Flow, Two stream Network, Observer Network, Virtual classroom

چکیده :

Online applications for virtual meetings have be- come handy for many fitness instructors these days. However, currently used applications have limitations in tracking clients’ performance, simultaneously and particularly for large number of participants. Thus, we proposed a fitness AI-assisted toolbox to detect fitness activities of athletes from steamed frames. We introduced a Magnitude flow method that computes the dense optical flow of each frame and provides details of human body poses during exercising. We introduced the two-stream convolutional neural network to only detect fitness actions from RGB frames and magnitude flows. We designed a light spatial network and a temporal network with a MobileNetV1 backbone and a Multi-layer perceptron (MLP) network as the classification head. Moreover, we introduced an Observer network as a score fusion approach to improve the final accuracy of detection and reduce the uncertainties between the two networks. We test our proposed method over a selected fitness class of the UCF dataset. We test the selected classes and acquired plenary results by considering its ability in detecting major fitness actions. Our efficient accuracies and confusion matrices describe that our proposed model is an effective solution to assist fitness instructors in tracking, analyzing and conveniently supervising a large number of participants.