Robust 3D Tracking of Multiple
Objects with Similar Appearances
1- Introduction
Detecting and tracking 3D
moving objects with a similar appearance is an essential step in video analysis
and is therefore widely used in machine vision systems such as surveillance,
traffic monitoring, vehicle navigation, human-computer interaction, interaction
human computer and robotic systems. They need to receive and process videos
received from their environment and finally analyze the behavior and events in
these videos. Since accuracy and speed are important factors in the optimal
performance of systems with low consumption time and high accuracy, it will
lead to an increase in the level of performance flexibility many challenges
such as changes in ambient light intensity, sudden change in the direction of
3D objects, the presence of different types of 3D objects in the field of view
of the camera, overlap, and so on.
2- Background
The 3D tracking
literature is evolving day by day due to its wide and significant applications
in the field of human security and comfort, so that it has emerged from the
shadow of military applications and has shown its application in modern systems
used by humans. According to the existing articles on the tracking of moving
three-dimensional objects, it can be said that the beginning of work on
tracking began approximately in the early 1980s and over the years its scope
and scope has increased. Among the comprehensive and complete work in the field
of tracking three-dimensional moving objects, which has been done by reviewing
and classifying tracking methods by ILMAZ (2006). The researchers tracked
moving three-dimensional objects using moving edge features. In this paper,
first, the fixed background edges are extracted by the canny edge algorithm,
and at each stage, the background fixed edges and the edges with partial
displacement are distinguished from the edges in the current frame by using the
difference of the moving edges from the fixed edges. Using the detected moving
edges, they performed the tracking operation.
3- Proposed Algorithm
In this algorithm, a new
method for simultaneous tracking of three-dimensional moving objects is
presented, which is not sensitive to the type of moving objects in the field of
view of the camera, the number of similar three-dimensional objects and their
different velocities. In this method, first using Sift corner points and
matching these points with KLT algorithm and based on the movement information
of key points related to moving objects, discover them and these points using
DBSCAN algorithm as three-dimensional moving objects in the tracking scene. We
label and at each step obtain these three-dimensional clusters or moving
objects on the received frames using the similarity of the object area and
averaging the brightness of the points belonging to that area.
3-1- Matching points with
KLT Algorithm
In the study method, we
use the Lucas-Kanade method to find the corresponding feature points between
two frames at moment t and t + 1. The main idea of the KLT method is based on
the three brightness constancy, temporal persistence and Spatial Coherence.
ccording to these three assumptions, the brightness of pixels belonging to an
object will be constant in successive frames, and the pixels in this sequence
will have small displacements, and also the points corresponding to a
particular pixel will have similar properties and displacements to each other.
In Figure (1), we show the corresponding feature points between two consecutive
frames whose correspondence is found between two frames using a blue line.
Fig
1: Corresponding feature points between two consecutive frames
3-2- Discovering Three-Dimensional Moving
Objects
In this step, using the
DBSCAN clustering algorithm, we divide the feature points associated with
similar 3D moving objects obtained in the previous step into different
clusters, each of which represents a three-dimensional moving object. The
purpose of this step is to discover the candidate areas for 3D moving objects
so that we can discover new 3D objects in the tracking stage based on the
similarity criteria. The basics of the DBSCAN algorithm are based on the fact
that clusters are areas of high-density data space separated by lower-density
regions. In this method, in order to estimate the distribution density of
points, two parameters of neighborhood radius (Eps) and the minimum points
required to form a cluster (MinPts) are used. The results of clustering
three-dimensional moving object related points for detecting moving objects can
be seen in Figure (2).
Fig
2: Results of clustering points related to three-dimensional moving objects to
detect moving objects
4- Analysis of the Proposed Algorithm
In the proposed method,
the Sift algorithm is used to extract the feature; because the points extracted
by this algorithm are resistant to scale changes and are less sensitive to
light and three-dimensional geometric changes. The purpose of using the DBSCAN
clustering algorithm is to separate three-dimensional moving objects from each
other and, in fact, to differentiate them. The results of the proposed method
are shown on three different videos in Figure (3), which shows the proposed
algorithm for tracking three-dimensional moving objects similar to the type of
objects, changes in speed and direction of movement, and the existence of
several three-dimensional moving objects simultaneously in the field of view of
the camera is not sensitive.
Fig
3: Results of tracking similar 3D moving objects
References
[1] Zhao Zhong, Zichen Yang,
Weitao Feng, Wei Wu, Yangyang Hu, and Cheng-lin Liu. Decision controller for object
tracking with deep reinforcement learning. IEEE Access, 2019.
[2] Weigang Lu, Zhiping Zhou, Lijuan Zhang,
and Guoqiang Zheng. Multi-target tracking by non-linear motion patterns based
on hierarchical network flows. Multimedia Systems, pages 1–12, 2019.
[3] Hilke Kieritz, Wolfgang Hubner, and
Michael Arens. Joint detection and online multi-object tracking. In Proceedings
of the IEEE Conference on Computer Vision and Pattern Recognition Workshops,
pages 1459–1467, 2018.
[4] R. Poppe, Vision-based human motion
analysis: an overview, Computer Vision and Image Understanding, vol. 108, no.
1, pp. 4- 18, Feb. 2007.
[5] A. Karamiani and N. Farajzadeh, Detecting
and tracking moving objects in video sequences using moving edge features, in
Proc. Scientific Cooperations International Workshops on Electrical and
Computer Engineering Subfields, pp. 88-92, Aug. 2014.
[6] R. Zhang and J. Ding, Object tracking and
detecting based on adaptive background subtraction, Procedia Engineering, vol.
29, no. 1, pp. 1351-1355, Feb. 2012.
[7] L. Nanni, A. Lumini, and S. Brahnam, Survey
on LBP based texture descriptors for image classification," Expert Systems
with Applications, vol. 39, no. 3, pp. 3634-3641, Feb. 2012.
[8] J. Ning, L. Zhang, D. Zhang, and C. Wu, Robust
object tracking using joint color-texture histogram, International J. of
Pattern Recognition and Artificial Intelligence, vol. 23, no. 07, pp. 1245-
1263, Nov. 2019.
[9] Dawei Zhao, Hao Fu, Liang Xiao, Tao Wu,
and Bin Dai. Multi-object tracking with correlation filter for autonomous
vehicle. Sensors, 18(7):2004, 2018.