Obtaining the most relevant 3d scene based on the given query.
Text-to-Scene RetrievalDifferent data types
Various Proposed Approaches
Primates Behavioural Analysis
Implemented for German Primate Center (DPZ)
Detection, Tracking, and Pose Estimation
Surveillance Video Synopsis and Analysis
Implemented for Iran's National Elites Foundation
Using OpenCv and deep learning techniques for saving essential events.
Two types of videos, including the traffic videos and surveillance videos(including pedestrians)have been
focused on in this project.
Developing some practical approaches for summarizing these types of videos
Implementing some analytical methods for anomaly detection
Using Hough Transform for lane detection
Using hierarchical clustering and dynamic time warping for directions clustering
Reconstructing Summarized Videos
Original Video (950 frames)Reconstructed Video (190 frames)
Different Analysis like obtaining the most common Trajectories
Fatigue Detection and Patient Assistant Tools using Eye key points
Using the eye key points to measure the level of fatigue by measuring number and Duration of eye
blinks
Eye tracking to ease the way of communication using a virtual keyboard
Action Recognition in Compressed Domain Video
Using compressed domain elements like Motion Vectors and Residuals for detecting events in a video.
Implementing a new approach by using a window for accumulating the residuals in order to reduce the
number of processed residual frames in compressed domain analysis.
Using one-dimensional convolutional layers for fast and efficient temporal pooling.
