Efficient Object Tracking Using Optimized K-means Segmentation and Radial Basis Function Neural Networks

Alireza Asvadi, Mohammadreza Karami, Yasser Baleghi
Babol University of Technology

Abstract:

In this paper, an efficient method for object tracking is proposed using Radial Basis Function Neural Networks. Optimized k-means color segmentation is employed for detecting an object in first frame. Next the pixel-based color features (R, G, B) from object is used for representing object color and color features from surrounding background is extracted and extended to develop an extended background model. The object and extended background color features are used to train Radial Basis Function Neural Network. The trained RBFNN is employed to detect object in subsequent frames while mean-shift procedure is used to track object location. The performance of the proposed tracker is tested with many video sequences. The proposed tracker is illustrated to be able to track object and successfully resolve the problems caused by the camera movement, rotation, shape deformation and 3D transformation of the target object. The proposed tracker is suitable for real-time object tracking due to its low computational complexity.

Paper:

1. Efficient Object Tracking Using Optimized K-means Segmentation and Radial Basis Function Neural Networks
Alireza Asvadi, Mohammadreza Karami, Yasser Baleghi
International Journal of Information and Communication Technology, Vol. 4, No. 1, pp. 29-39, December 2011.
(This paper is a substantial extension of our MVIP2011 paper)
[link to pdf]

2.  Improved Object Tracking Using Radial Basis Function Neural Networks
Alireza Asvadi, , Mohammadreza Karami, Yasser Baleghi, Hossein Seyyedi
in Proceedings of 7th Iranian Machine Vision and Image Processing (MVIP2011), Tehran, Iran, November 2011.
[IEEE indexed] (Selected as a top 16 English papers of MVIP2011)
[PDF] [Poster]

Code:

The code runs on Windows XP with MATLAB R2011a.
a test sequence “test.avi” is included so you can simply run Demo.m
[Code]

Result:

Tracking result for proposed method (dashed red), Mean shift method (dash-dot blue) and ground truth (solid green). Binary images show the object silhouette obtained by the proposed method.

Fig8

Fig11

Fig14