Learning Spatial and Temporal Extents of Human Actions for Action Detection

Project Discription

The world is being digitalized so it can be easily seen that the digital data is an important role player in today’s world. The prospectus of computer vision has so many different applications for further elaboration of research. Out of these so many applications, video tracking is one of the interesting as well as highly important application of research. Video tracking basically is a combination of two sequential processing activities. It consists of object detection and the recognition of action of that particular detected object.  In the detection system of proposed method automatically identifies the action performed by the human like running, moving, kicking, and etc. Such type of work can be used to monitor the sensitive area where the continuous monitoring is required. If there is any kind of unfair activity, system may generate alarm. There are so many researches and so many researchers, who are continuously working on video object detection issue. The proposed method in this paper works on video objects detection with action reorganization in video. 

Whole work is explained in figure  where block diagram shows the extraction of feature from the input training video then pass these video feature in the neural network for training.

As video is the collection of images which is called as frame. Here these collections of images are display in so fast sequence that human eyes could not judge that it actually see one image at a time. As contents of the consecutive frames are mostly same but change in object position is new information of the frame. So reading of video means conversion of video in sequence of frames of RGB format.

Project Sample Image