Please use this identifier to cite or link to this item:
Title: Challenges in action recognition in videos
Authors: Lee, Liang Cheng
Keywords: DRNTU::Engineering
Issue Date: 2016
Abstract: With the discovery of action recognition and object detection in computer vision, there has been an increase in research interest among this field of study as many different methodologies and techniques are being explored, experimented, analyzed and proposed to improve the efficiency and robustness of the existing ability of the program to recognize or detect an object. However, there are certain challenges that will always pose as an obstacle which will affect the accuracy of the program. Several well-known challenges such as background cluttering, view point, camera motion and occlusions often posed as an obstacle in action recognition. Due to the complexity in nature of these challenges especially occlusion, which makes detection in video difficult, much attention and research focus is required to address all this challenges. To depict the challenges faced in action recognition in videos, a relatively new yet efficient technique known as dense trajectory will be adopted while utilizing several different types of dataset to determine the output accuracy, so as to determine and study the effects that these challenges could cause. In this report, the dense trajectory method will compute the features using dense sampling and optical flow field to extract dense trajectory followed by concatenation of computed features by 4 descriptors to determine its robustness in handling challenges of action recognition. On top of basic action recognition testing, different types of test known as view base testing and occlusion effect testing was being carried out as well. These challenges of action recognition was then identified and analyze based on the eventual accuracy attain. The next part is to adopt the second method known as log-covariance matrix method with motion context to serve as a comparative study with dense trajectory method in terms of robustness in handling these challenges. Based on the eventual accuracy attained from these two methods, it can be shown that dense trajectory method outperforms log-covariance matrix with motion context method.
Rights: Nanyang Technological University
Fulltext Permission: restricted
Fulltext Availability: With Fulltext
Appears in Collections:EEE Student Reports (FYP/IA/PA/PI)

Files in This Item:
File Description SizeFormat 
Lee Liang Cheng - U1322641C - FYP Final Report.pdf
  Restricted Access
Final Year Project Report4.63 MBAdobe PDFView/Open

Page view(s)

Updated on Jun 21, 2021


Updated on Jun 21, 2021

Google ScholarTM


Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.