SMARTEYE for human posture and activity monitoring Part 1
Soh, Pei Fang
Date of Issue2016-11-07
School of Computer Engineering
The objective of this project is to research on methods to implement human detection, pose estimation, and activity recognition from pre-recorded 2-dimensional video sequences. A combination of techniques – the Edge Detection, Frame Differencing, Gaussian Mixture Models and Optical Flow – was used to achieve human detection. A bounding box is drawn onto the image region where the human is assumed to be by the human detection algorithm. The values of the bounding box were fed into the Convolutional Pose Machines to process a segment of the image for joint locations. The joint locations are then joined to form a skeletal figure, which can be used for human pose estimation. Based on a Fuzzy Inference System, the human activity being performed is determined. The experimental results of the implementation had shown improvements in various aspects of the human detection and activity recognition when compared against ground truth and prior work.
Final Year Project (FYP)
Nanyang Technological University