Classification of Human Actions in Outdoor Images Using Hybrid Deep Learning Methods

In this paper, a hybrid deep learning approach is proposed for automatic classification of human actions in outdoor images. Firstly, the detection of the humans in the image taken from the outdoor environment is provided. For this purpose, YOLO, a pre-trained deep object detection tool widely used in the literature, is used. The Google street view platform was preferred to acquire outdoor images. Action classes were created for the detected humans. These action classes; walking to the right, walking to the left, standing and sitting. Thus, a comprehensive dataset has been created for humans identified from the outdoor images. A convolutional neural network (CNN) model is designed for the automatic recognition of classes. With this designed CNN model, automatic recognition of the human actions is provided. This trained model is used in a hybrid manner with the YOLO object detection system to automatically recognize the human actions in the input images.Within the scope of the article, a four-class dataset was created and performance evaluations of the proposed system were made.