DESIGN OF OBJECT TRACKING ALGORITHM BASED ON MEAN-SHIFT WITH OPTIMIZATION USING SIMULATED ANNEALING

• Main Author: Ardiansyah, Aris
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/36653
• Institution: Institut Teknologi Bandung

The mean-shift algorithm is a simple and fast object tracking algorithm, this advantage is obtained because the mean-shift algorithm does not take all image pixels as input for algorithms. This algorithm only considers the pixels in object area being tracked, so that the computational load of this algorithm is minimal. The mean-shift algorithm is suitable for use in real-time conditions in terms of execution time. The use of histograms also causes this algorithm to be relatively resistant to rotation and change in the size of the object being tracked. However, the resistance of this algorithm to lighting conditions and changes is not optimal. The study aims to improve the performance of algorithms in lighting change conditions while reducing algorithm processing time. Some of the treatments proposed by the author include: the use of Hue color components replacing RGB color model, the use of sampling techniques to reduce histogram iterations on formation and direction estimation, optimization of candidate search object locations using simulated annealing algorithm, adding tolerance to search area for optimal object positioning and using area based weighting instead of kernel Epanechnikov which is based on the Euclidean distance that used in the meanshift algorithm. The test results using 999 frames of video images with conclusion using one tail ttest with two independent sample groups showed that the average performance of the proposed algorithm is significantly better than the mean-shift algorithm in terms of lighting resistance and time to process per video frame.