• 
• 
• 
• 
• 
• 

Low-Frame-Rate Tracking

As a part of the Physical Security SK-Pro, we have developed a robust, computationally efficient multi-object tracking method that runs for low-frame-rates as well as fast moving object scenarios.

Background & Objective:  Advanced video surveillance systems are assembled from a large number of cameras. It is desired to achieve real-time tracking performance while keeping the hardware costs on a minimum level in such systems. Therefore, it is necessary to process the vast amount of constantly streaming channels of video on a single CPU at the same time. However, most existing tracking approaches presume they can consume all the available processing power to track objects in a single sequence.
     One solution to the challenging problem of processing of multiple video sequences on the same CPU is to sample every input video such that the number of frames per second is decreased proportional to the number of sequences. However, due to the decrease of the frame rate, the tracking algorithm receives video frames at a lower temporal resolution, which causes the objects to appear reciprocally much faster than to the original sequence. Since object movements are usually large and unpredictable, existing tracking approaches fail to detect the target objects in the current frame.

Technical Discussion:  We present an object tracking algorithm for low-frame-rate applications. We assign multiple kernels centered on high motion areas. We improve the convergence properties of the mean-shift by integrating two additional likelihood terms. Unlike the existing approaches, the proposed algorithm enables tracking of moving objects at lower temporal resolutions as much as 1-fps frame rate without sacrificing the robustness and accuracy. Therefore, it can process multiple videos at the same time on a single processor. Note that, the low frame rate constraint corresponds to the fast motion of the moving objects. Thus, the proposed method is capable of tracking fast objects even in the original frame rates.

Publications:
Tuzel, O.; Porikli, F.; Meer, P., "A Bayesian Approach to Background Modeling", IEEE Workshop on Machine Vision for Intelligent Vehicles (MVIV), ISSN: 1063-6919, Vol. 3, pp. 58, June 2005 (IEEE Xplore, TR2005-033)

Porikli, F.; Wren, C.R., "Change Detection by Frequency Decomposition: Wave-Back", Workshop on Image Analysis for Multimedia Interactive Services, April 2005 (WIAMIS 2005, TR2005-034)

Porikli, F.; Tuzel, O., "Object Tracking in Low-Frame-Rate Video", SPIE Image and Video Communications and Processing, Vol. 5685, pp. 72-79, March 2005 (SPIE Publications, TR2005-013)

Porikli, F., "Multiplicative Background-Foreground Estimation Under Uncontrolled Illumination using Intrinsic Images", IEEE Computer Society Workshop on Motion and Video Computing, January 2005 (IEEE MOTION 2005, TR2005-012)

Porikli, F.M., "Road Extraction by Point-wise Gaussian Models", SPIE Algorithms and Technologies for Multispectral, Hyperspectral and Ultraspectral Imagery IX, Vol. 5093, pp. 758-764, September 2003 (SPIE Proceedings, TR2003-059)

Porikli, F.M.; Divakaran, A., "Multi-Camera Calibration, Object Tracking and Query Generation", IEEE International Conference on Multimedia and Expo (ICME), Vol. 1, pp. 653-656, July 2003 (IEEE Xplore, TR2003-100)

Technology Area:  Computer Vision

Modification Date:  June 16, 2006