LSTM based video stabilization for object tracking

Chenyi Yang; Yuqing He; Danfeng Zhang

doi:10.1117/12.2606941

24 November 2021 LSTM based video stabilization for object tracking

Chenyi Yang, Yuqing He, Danfeng Zhang

Proceedings Volume 12065, AOPC 2021: Optical Sensing and Imaging Technology; 120653D (2021) https://doi.org/10.1117/12.2606941
Event: Applied Optics and Photonics China 2021, 2021, Beijing, China

Abstract

The object tracking accuracy may be decreased because of the camera jitter, making it extremely hard for object tracking and trajectory analyzation. To achieve accurate video stabilization, the movement of camera can be analyzed and predicted based on the previous camera jitter sequence. In the area of sequence prediction, the long-short term memory (LSTM) network shows the potential in sequence forecasting, here we use LSTM network in camera jitter prediction and video stabilization. In this paper, we propose a video stabilization algorithm based on multi-region grey projection method and LSTM encoder-decoder network. Our algorithm calculates the motion of the camera through the gray projection of four areas in each frame, then filters out the main movement direction and jitter of the camera. The LSTM encoder-decoder network receives the camera jitter sequence, predicts the camera jitter then stabilizes the video. We to verify the performance of the proposed video stabilization method. We tested the proposed video stabilization algorithm on the jitter videos, which is made by the VisDrone dataset video modified with our recorded camera jitter. Experimental results demonstrate that the proposed method can achieve the video stabilization in real time, and increase the accuracy of object tracking and trajectory analyzation.

Citation Download Citation

Chenyi Yang, Yuqing He, and Danfeng Zhang "LSTM based video stabilization for object tracking", Proc. SPIE 12065, AOPC 2021: Optical Sensing and Imaging Technology, 120653D (24 November 2021); https://doi.org/10.1117/12.2606941

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available