Fast Hand Tracking with Deep CNN

The aim of this project is to assess the state of the art in object tracking in a video, with an emphasis on the accuracy and the robustness of tracking algorithms. As there has been little conceptual consensus between most of the methods, we have sought to describe the state of the art at the other end: the data. We use the MOT dataset which composes of a set of real-life scenarios as diverse as we could consider and have recorded the performance of all selected trackers on them. We aim to group methods of tracking on the basis of their experimental performance.

About the Instructor

Mr. Joseph R Issac is a PhD Scholar at the Department of Computer Science and Engineering, IIT Madras.

Joseph is the first student from Jeppiaar Engineering college to join for Direct Ph. D. in IITM. He secured All India Rank 883 in GATE 2015. Currently he is working on Sensor fusion and tracking of hands in 3D space with humanistic filter.

LinkedIN: https://www.linkedin.com/in/joehrisaac/

 

About the Project 

Many trackers have been proposed in the literature which have different ad- vantages and disadvantages during the last two decades. Object tracking is a well-established but unsolved problem since there are many factors which affect the performance of object tracking. An ideal tracker must track the object despite changes in the appearance, motion, occlusion, image quality etc. Many trackers have been proposed solving one or more of these issues. However, the performance of proposed trackers have been evaluated with different videos of varying scenarios. In this report, we aim to evaluate trackers systematically and experimentally on the MOT video dataset. We selected a set of five trackers to include a wide variety of algorithms often cited in literature, appearing from 2015 to 2017 for which the code was publicly available. Multiple object tracking (MOT) is a problem in computer vision which has been well studied. The main objective of MOT is to track multiple objects simultaneously and with good accuracy and efficiency. Each object tracked must be uniquely identified and each individual trajectory must be tracked. One common example of MOT is pedestrian tracking, where the goal is to track the individual people in a particular scenario. In this review, we mainly focus on the research on pedestrian tracking. The underlying reason for this specification is that compared to other common objects in our environment, pedestrians are typical non-rigid objects, which is an ideal example to study the MOT problem.

Given the wide variety of aspects in tracking circumstances, and the wide variety of tracking methods, the evaluation of these methods is limited and each method was evaluated with different datasets. The aim of this project is to assess the state of the art in object tracking in a video, with an emphasis on the accuracy and the robustness of tracking algorithms. As there has been little conceptual consensus between most of the methods, we have sought to describe the state of the art at the other end: the data. We use the MOT dataset which composes of a set of real-life scenarios as diverse as we could consider and have recorded the performance of all selected trackers on them. We aim to group methods of tracking on the basis of their experimental performance.

Prerequisites 

  • Previous Experience with Python, Tensorflow and Computer Vision
  • A PC/Laptop with a Python IDE and a GPU (only recommended)
  • Knowledge about Python, Tensorflow, OpenCV

Week 1

1
Introduction

Week 2

1
Topic will be decided after completing Previous sessions

Week 3

1
Topic will be decided after completing Previous sessions

Week 4

1
Topic will be decided after completing Previous sessions

Week 5

1
Topic will be decided after completing Previous sessions

Week 6

1
Topic will be decided after completing Previous sessions

Week 7

1
Topic will be decided after completing Previous sessions

Week 8

1
Conclusion
Get course
30-Day Money-Back Guarantee

Includes

16 LIVE Sessions
Full lifetime access
Access on mobile and TV