حجت عسگريان دهكردي

Recently still image-based human action recognition has become an active research topic in computer vision and pattern recognition. It focuses on identifying a person's action or behavior from a single image. Unlike the traditional action recognition approaches, where videos or image sequences are used, a still image contains no temporal information for action characterization. Therefore, it is demanding to develop efficient methods for still image-based action recognition. Deep learning methods such as convolutional neural networks (CNNs) have recently emerged as a dominant paradigm for machine learning in a variety of domains and they can deliver highly accurate classification results when provided with large enough data sets and respective labels. However, acquiring a suitably large data set for training DCNN is often a significant challenge. This is a major issue in the action recognition domain, where we lack large enough labeled data. Using CNNs along with limited labeled data can be problematic, as this leads to extensive overfitting. In this research, we propose a two-phase multi-expert classification method for human action recognition by means of super-class learning and without any extra information. Specifically, a coarse-grained phase selects the most relevant fine-grained experts. Then, the fine-grained experts encode the intricate details within each super-class so that the inter-class variation increases. In the proposed approach, to choose the best configuration for each super-class and characterize inter-class dependency between different action classes, we propose a novel Graph-Based Class selection (GCS) algorithm. Extensive experimental eva‎luations are conducted on various public human action recognition datasets, including Stanford40, Pascal VOC 2012 Action, and BU101+ dataset. The experimental results demonstrate that the proposed method yields promising improvements. To be more specific, on Sanford40, Pascal VOC 2012 Action, and BU101+ benchmarks, the proposed approach outperforms the state-of-the-art studies by 0.41%, 0.66%, and 2.11 % mAP with much less computational cost and without any auxiliary annotation information such as human and object bounding boxes.

شناسايي فعاليت انسان، يادگيري عميق، شبكه¬هاي عصبي كانولوشني عميق، انتقال يادگيري، يادگيري گروهي

human action recognition, deep learning, deep convolutional neural network, transfer learning, ensemble learning

Hojat Asgariandehkordi

Shahriar Baradaran Shokouhi