Salient Closed Boundary Tracking

This code is implemeted for the paper
"Real-Time Salient Closed Boundary Tracking via Line Segments Perceptual Grouping, Xuebin Qin, Shida He, Camilo Perez Quintero, Abhineet Singh, Masood Dehghan and Martin Jagersand."
which is submited to IROS 2017.

ABSTRACT

This paper presents a novel real-time method for tracking salient closed boundaries from video image sequences. This method operates on a set of straight line segments that are produced by line detection. The tracking scheme is coherently integrated into a perceptual grouping framework in which the visual tracking problem is tackled by identifying a subset of these line segments and connecting them sequentially to form a closed boundary with the largest saliency and a certain similarity to the previous one. Specifically, we define a new tracking criteria which combines a grouping cost and an area similarity constraint. This criteria makes the resulting boundary tracking more robust to local minima. To achieve real-time tracking performance, we use Delaunay Triangulation to build a graph model with the detected line segments and then reduce the tracking problem to finding the optimal cycle in this graph. This is solved by our newly proposed closed boundary candidates searching algorithm called "Bidirectional Shortest Path (BDSP)". The efficiency and robustness of the proposed method are tested on real video sequences as well as during a robot arm pouring experiment.

The demo video can be found https://youtu.be/RXjD0yHkukI.

The dataset is published https://github.com/NathanUA/SalientClosedBoundaryTrackingDataset.

USED LIBRARIES

This implementation is based on Opencv 2.4.9(3.1.0), Boost 1.63.0 and ubuntu 14.04 64 bit.

The line detector used here is EDLines(EDLines.a) which is proposed in
"C. Akinlar and C. Topal, EDLines: A real-time line segment detector with a false detection control, Pattern Recognition Letters, vol. 32, no. 13, pp. 1633-1642, 2011."
and can be downloaded from http://ceng.anadolu.edu.tr/CV/downloads/downloads.aspx. We include the lib in the root directory. It is worthynote that we are using the 64 bit ubuntu version of EDLines. We sugggest to test our algorithm on a ubuntu 64bit OS.

The RCC based tracker is adapted from the method developed in
"J. S. Stahl and S. Wang, “Edge grouping combining boundary and region information,” IEEE Trans. Image Processing, vol. 16, no. 10, pp. 2590–2606, 2007."
We download the code from https://cse.sc.edu/~songwang/software.html.

QUICKSTART

Git clone or Download the zip file and unzip it.
Go to the root path
"make"
then, use command "./SalientBoundaryTracking video_proc gp_flag video_path" to run the code.
For example, read video from .avi file:
"./SalientBoundaryTracking 1 1 ./TEST_VIDEO.avi"
or read video from webcam:
"./SalientBoundaryTracking 0 1"

video_proc: "0": read file from your webcam, "1": read video from a .avi file
gp_flag: "0": use RRC adapted method to track, "1": use our BDSP to track
video_path: specify the video sequence to be loaded. If video_proc == 0, it will be neglected.

TRACKING CONTROL

(1). Click mouse left button to initialize a
　　polygon in "Display" window.
　　When read video from a webcam,
　　you have to press key "i" before
　　initialization.
(2). Click mouse middle button to finish initialization.
(3). Press key "space" to run tracking continuously or
　　press key "s" to run tracking frame by frame.
　　Key "r" can recover the
　　continuously tracking model)."Esc" will quit the tracking.
(4). Tracking output are two or three files:
　　"TRACKING_RESULTS.txt"
　　"RECORDED_RESULTS.avi"
　　"RECORDED_VIDEO.avi"(if video_proc==0).

CONTACTS

Xuebin Qin
Department of Computing Science
University of Alberta
Edmonton, AB, Canada, T6G 2E8

Email:
[email protected]
[email protected]