Object detection for a non-static (Panning and Tilting) camera has been the centre of interests of various applications. In this paper, we present a novel real-time object detection approach that can efficiently handle the problem of a moving camera that performs panning and tilting motions to incorporate a wide field of view for more security and a wider angle of view. We present a set of possible solutions where the optimal is implemented on an embedded Texas Instrument DM6437 DSP platform which is suitable for video processing. In the presence of a non-static camera, classical object detection algorithms cannot be used. Therefore our idea consists of two main stages: The first stage attempts to align the two successive frames of the video by using a method based on 2D -correlation (Sum of absolutes differences (SAD)); the second stage involves one of the classical background subtraction algorithms. The system was successively simulated in Matlab environment and then implemented in DSP platform.
[1]
Yiming Ye,et al.
Tracking a Person with Pre-recorded Image Database and a Pan, Tilt, and Zoom Camera
,
1998,
ACCV.
[2]
Sergio A. Velastin,et al.
How close are we to solving the problem of automated visual surveillance?
,
2008,
Machine Vision and Applications.
[3]
W. Eric L. Grimson,et al.
Adaptive background mixture models for real-time tracking
,
1999,
Proceedings. 1999 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (Cat. No PR00149).
[4]
Bang Jun Lei,et al.
Real-time outdoor video surveillance with robust foreground extraction and object tracking via multi-state transition management
,
2006,
Pattern Recognit. Lett..