Professional Documents
Culture Documents
A High Performance Object Tracking Technique With An Adaptive Search Method in Surveillance System
A High Performance Object Tracking Technique With An Adaptive Search Method in Surveillance System
A High Performance Object Tracking Technique With An Adaptive Search Method in Surveillance System
Abstract— In the video scene, the technique on tracking When people are interacted with each other or the occlusion
multiple targets, such as tracking group of people through is caused by other foreground objects, the tracking problem is
occlusion, is still challenging. In this paper, we present an occurred with occlusions, One solution for occlusions problem
algorithm for multiple targets tracking system. We discuss the is utilizing multiple views to compensate for the insufficiency
behavior of the moving objects with adaptive search method. visibility of a single view [1]. However multiple views are not
Several cases are classified, including the no match case, only-one always available for some applications. A popular solution
match case, split case and occlusion case respectively. With this strategy for detecting and tracking multiple people in
system, we can track the moving people in successive frame by surveillance situations is the use of probabilistic appearance
object boundary box and velocity without color cues or
models [2]-[4], which use probabilistic models to characterize
appearance model. Even though people are interacted with each
these people as the appearance and localization model to track
other or the occlusion is caused by other foreground objects, the
proposed algorithm can still perform well. Furthermore we the target. Some method performs segmented objects and
consider the people movement with respect to the distance with explicit tracking using Kalman filter [5], [6]. W4 in [7] is a
the camera as an adaptive search range to deal with the condition. tracking system based on shape analysis for the surveillance.
As the foreground is similar to the background, the proposed These methods always have to extract the feature or establish
algorithm can still solve the problem on the detection error. model from object and thus take complex computation.
To be applied in the surveillance system, in this paper we
Keywords—object tracking; surveillance system; object
grouping;
present a high performance object tracking technique to deal
with the occlusion problem. We utilize the object boundary
box, and record the object velocity. By this information we can
I. INTRODUCTION preserve the identity of an object across a sequence of frame.
In a surveillance system, the system transmits video, This paper is organized as follows. Section II presents the
stationary cameras extracting, to central server by wires or overall algorithm. Some experimental results are shown in
wireless. The traditional surveillance networks rely on labor Section III. Conclusions are provided in Section IV finally.
diagnose the activities on monitor. But there is a limit on the
amount of monitor which human can handle. Human is also II. OVERALL OF THE PROPOSED SYSTEM
easily distracted by the ambient environments changing. That
causes the important events easily missing or erroneous The main goal of the proposed method is to track multiple
judgment. targets. At first, Multi-model Background Maintenance (MBM)
algorithm is used to extract moving blob [8]. Based on the
The intelligent surveillance system means that it is capable previous research works [9], we use the approach with a
of automatically multiple information extraction and analysis. combination of run-length encoding (RLC) and object labeling
Following the development of intelligent surveillance systems, of algorithms. Labeling procedure gives a label value on each
the systems observably reduce the cost of human resource. In object. The small area objects will be considered as noise. The
this kind of system, the object tracking technique has received proposed tracking algorithm is shown in Fig. 1. Object
more and more attention in recent years, due to the drive from
grouping and object tracking are discussed later.
many emerging applications. To match and detect multiple
foreground objects between consecutive frames, it can be used A. Object grouping
by different features of object such as shape, velocity, color, When the pixel characteristics of a foreground object are
texture. Tracking is the process to track the objects over the too similar to the background, it is hard to discern between
time by locating its position in every frame of the video foreground and background. For example, when a person is
Tracking single human or a small group of humans for passing wearing a cloth, this situation presents a similar color or pixel
is always existed with occlusion. Moreover, in video scene illumination to the background. It causes the foreground
tracking group of people, which have been through occlusion, detection fault. As shown in Fig. 2(a), the foreground objects
is still challenging. are incorrectly considered as background pixels. Then one
2FFOXVLRQFDVH
(a) (b) (c) recorded from its values in the previous frames as
1 n
Fig. 2: An example of object grouping: (a) foreground segmentation result,
Vo(i ) = ¦ Vo(t − m ) (i )
(b) object blob without grouping and (c) object grouping result. n m =1
Object tracking is aimed at tracking the identity of an object
Vo(i )
across a successive of frames with small groups of people where is the estimated velocity of the i-th object in the
moving together or interacting with each other. In these cases, image frame at time t. There are two pairs of object matched
individual people are not visually isolated, and could be in Fig. 5. In this case there is a matching between object blob
partially or total occluded by other people. To solve this Cnt and object blob Pmt −1 . Here we adopt the object Pmt −1 label
354
Fig. 4: Case of on match pair Fig. 8: Case of occlusion problem.
Fig. 9: Utilizing the object label value and velocity in previous frame.
Fig. 5: Case of only-one match pairs.
355
Fig.11: Tracking result for people is occulted by digital number case.
Fig. 12: Tracking result for people is interacting with each other case.
Fig. 13. Tracking result for PETS2010 dataset view 007 by fixed threshold..
Fig. 14: Tracking result for PETS2010 dataset view 007 by proposed method.
the relationship on the distance with the camera and the people Trans. Circuits and systems for video technology, vol. 16, no. 12,
are in front of camera, people have more amount of movement. December 2006.
This derives the adaptive search range in our algorithm to deal [6] A. Amer, “Voting-based simultaneous tracking of multi-ple video
objects”, IEEE Trans. Circuits and systems for video technology, vol.
with the occlusion condition well. Moreover the object 15, no. 11, November 2005
grouping method also can solve the problem of foreground [7] I. Haritaoglu, D. Harwood, and L. S. Davis,” W4: real-time
detection fault when foreground is similar to background. surveillance of people and their activities” IEEE Trans. Pattern Anal
Simulation results show that more than 94% of these Machine intell., vol. 22, no. 8, August 2000
correspondence match accuracy is achieved. [8] T.-H. Tsai, W.-T. Sheu and C.-Y. Lin, "Foreground Object Detection
based on Multi-model Background Maintenance ", IEEE International
Symposium on Multimedia, ISM ’07, Taiwan , 2007.
REFERENCES
[9] T.-H. Tsai, C.-C. Huang and C.-H. Chang,”Design on Wireless
[1] M. Shah and S. Khan. “Tracking multiple occluding people by localizing Intelligent Sensor Network on Cloud Computing System for Smart
on multiple scene planes.” IEEE Transactions on Pattern Analysis and Home”, Sixth International Conference on Wireless & Mobile Networks,
Machine Intelligence, 31–3:505–519, 2009. India ,2014
[2] A. Senior, A. Hampapur, Y. Tian, L. Brown, S. Pankanti, and R. Bolle.” [10] Yilmaz, A., Javed, O., and Shah, M. 2006. “Object tracking: A survey”.
Appearance models for occlusions handling. “In IEEE International ACM Comput. Surv. 38, 4, Article 13,December 2006
Workshop on Performance Evaluation of Tracking and Surveillance, [11] Conte, D.; Foggia, P.; Percannella, G.; Vento, M., "Performance
2001. Evaluation of a People Tracking System on PETS2009 Database,"
[3] A. Elgammal and L. Davis. “Probabilistic framework for segmenting Advanced Video and Signal Based Surveillance (AVSS), 2010 Seventh
people under occlusion.” In IEEE International Conference on Computer IEEE International Conference on , vol., no., pp.119,126, Aug. 29 2010-
Vision, 2001. Sept. 1 2010
[4] T. Zhao and R. Nevatia. “Tracking multiple humans in complex [12] PETS2010“, In Conjunction with IEEE Computer Society (PAMI TC)
situations.” IEEE Transactions on Pattern Analysis and Machine and IEEE Signal Processing Society (IVMSP TC) Boston, US - 29th
Intelligence, 26–9:1208–1221, 2004. August 2010
[5] P. Kumar, S. Ranganath, K. Sengupta, and H. Weimin, “Cooperative
Multitarget tracking with efficient split and merge Handling”, IEEE
356