Portal de Periódicos da CAPES

Indian Classical Dance Forms Classification Using Transfer Learning

2022; Springer International Publishing; Linguagem: Inglês

10.1007/978-981-19-3391-2_18

2367-4512

Challapalli Jhansi Rani, Nagaraju Devarakonda,

Gait Recognition and Analysis

Human activity analysis is useful in a variety of domains, including video surveillance, biometrics, and home health monitoring systems. In computer vision field, extraction and recognition of complex human movements from images/videos are a great complex task. In this present work, we propose the Indian classical dances (ICD) classification using the concept of transfer learning. ICD form is a combination of gesticulation of all body parts. It comes in a variety of shapes and sizes, but the most common features include single/double hand mudras, eye movement, legs alignment, hip movements, facial expressions, and legs posture. Each dance has its own gestures and clothes worn by the dancers. India classical dances are categorized into 8 categories. In this work, we used the dataset consists of eight dance classes includes Bharatnatyam, odissi, manipuri, kuchipudi, mohiniyattam, sattriya, kathakali, and kathak. Those images were collected from Internet. While image processing using CNN model training with less data does not give accurate result that leads to over-fitting problem. To overcome this problem, we propose a concept, transfer learning by this use the knowledge that was learned from some problem that can be applied to solve the problem related to the target task. It reduces both time and space complexity. In our proposed work, we use a pre-trained model VGG16. It results high accuracy of 85.4% compared to earlier methods.