Dense Flow-Based Video Object
Segmentation in Dynamic Scenario
Arati Kushwaha, Om Prakash, Rajneesh Kumar Srivastava
and Ashish Khare
Abstract Segmenting object from a moving camera is a challenging task due to
varying background. When camera and object both are moving, then object segmen-
tation becomes more difficult and challenging in video segmentation. In this paper,
we introduce an efficient approach to segment object in moving camera scenario. In
this work, first step is to stabilize the consecutive frame changes by the global cam-
era motion and then to model the background, non-panoramic background modeling
technique is used. For moving pixel identification of object, a motion-based approach
is used to resolve the problem of wrong classification of motionless background pixel
as foreground pixel. Motion vector has been constructed using dense flow to detect
moving pixels. The quantitative performance of the proposed method has been cal-
culated and compared with the other state-of-the-art methods using four measures,
such as average difference (AD), structural content (SC), Jaccard coefficients (JC),
and mean squared error (MSE).
Keywords Object segmentation · Moving camera · Dense flow
Non-panoramic model · Background modeling
A. Kushwaha · R. K. Srivastava · A. Khare (B )
Department of Electronics and Communication,
University of Allahabad, Allahabad, Uttar Pradesh, India
e-mail: ashishkhare@hotmail.com
A. Kushwaha
e-mail: aratikushwaha.jk@gmail.com
R. K. Srivastava
e-mail: rkumarsau@gmail.com
O. Prakash
Department of Computer Science and Engineering,
Nirma University, Ahmedabad, India
e-mail: au.omprakash@gmail.com
© Springer Nature Singapore Pte Ltd. 2019
A. Khare et al. (eds.), Recent Trends in Communication, Computing,
and Electronics, Lecture Notes in Electrical Engineering 524,
https://doi.org/10.1007/978-981-13-2685-1_26
271