Multi-Class Confidence Detection Using Deep Learning Approach

The advancement of both the fields Computer Vision (CV) and Artificial Neural Networks (ANNs) has enabled development effective automatic systems for analyzing human behavior. It is possible to recognize gestures, which are frequently used by people communicate information non-verbally, studying hand movements. So, main contribution this research collected dataset, taken from open-source videos relevant subjects that contain actions depict confidence levels. dataset contains high-quality frames with minimal bias less noise. Secondly, we have chosen domain determination during social issues such as interviews, discussions, or criminal investigations. Thirdly, proposed model a combination two high-performing models, i.e., CNN (GoogLeNet) LSTM. GoogLeNet state-of-the-art architecture detection gesture recognition. LSTM prevents loss keeping temporal data. So these outperformed training testing process. This study presents method different categories Self-Efficacy performing multi-class classification based on current situation movements using visual data processing feature extraction. pre-processes sequence images scenarios, including humans, their quality extracted. These then processed extract analyze features regarding body joints position classify them into four classes related efficacy, confidence, cooperation, confusion, uncomfortable. extracted framework customized Convolutional Network (CNN) layers Long Short-Term Memory (LSTM) extraction classification. Remarkable results been achieved representing 90.48% accuracy recognition gestures through deep learning approaches.