Visual tracking using correlation techniques : algorithms, analysis, and pipeline architecture implementation
In this thesis, a correspondence based tracking algorithm was detailed. This tracking algorithm tracked a region of an image located on an object. It used corrospondence between the training region and the current region to locate the object in the current image. Since manipulations of the object could dramatically change the geometry of the training region, the training region was updated while the tracking algorithm ran. Six steps were takin in creating the tracking algorithm. These were to
1. Define the issues associated with tracking.
2. Examine seven different tracking algorithms and choose a correspondence based tracking algorithm.
3. Introduce four correspondence algorithms to be compared.
4. Compare these four algorithms for accuracy and speed.
5. Choose the Minimum Noise and Correlation Energy (MINACE) filter for correspondence.
6. Build and evaluate a system that uses this tracking algorithm.
The tracking algorithm was implemented on a VME rack equipped with a Datacube MV20 pipeline processor and a Motorola 68030 processor. A data update rate of 15hz was achieved.
Thesis94.E48.pdf
13.87 MB
Adobe PDF
20ab2ce0da9a65a79664b3c3dabc4584