Midrange exploration exploitation searching particle swarm optimization with HSV-template matching for crowded environment object tracking

Particle Swarm Optimization (PSO) has demonstrated its effectiveness in solving the optimization problems. Nevertheless, the PSO algorithm still has the limitation in finding the optimum solution. This is due to the lack of exploration and exploitation of the particle throughout the search space. Th...

Full description

Saved in:
Bibliographic Details
Main Author: Nurul Izzatie Husna, Muhamad Fauzi
Format: Thesis
Language:English
Published: 2023
Subjects:
Online Access:http://umpir.ump.edu.my/id/eprint/41486/1/ir.NURUL%20IZZATIE%20HUSNA%20-PCC17011.pdf
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Particle Swarm Optimization (PSO) has demonstrated its effectiveness in solving the optimization problems. Nevertheless, the PSO algorithm still has the limitation in finding the optimum solution. This is due to the lack of exploration and exploitation of the particle throughout the search space. This problem may also cause the premature convergence, the inability to escape the local optima, and has a lack of self-adaptation in their performance. Therefore, a new variant of PSO called Midrange Exploration Exploitation Searching Particle Swarm Optimization (MEESPSO) was proposed to overcome these drawbacks. In this algorithm, the worst particle will be relocating to a new position to ensure the concept of exploration and exploitation remains in the search space. This is the way to avoid the particles from being trapped in local optima and exploit in a suboptimal solution. The concept of exploration will continue when the particle is relocated to a new position. In addition, to evaluate the performance of MEESPSO, we conducted the experiment on 12 benchmark functions. Meanwhile, for the dynamic environment, the method of MEESPSO with Hue, Saturation, Value (HSV)-template matching was proposed to improve the accuracy and precision of object tracking. Based on 12 benchmarks functions, the result shows a slightly better performance in term of convergence, consistency and error rate compared to another algorithm. The experiment for object tracking was conducted in the PETS09 and MOT20 datasets in a crowded environment with occlusion, similar appearance, and deformation challenges. The result demonstrated that the tracking performance of the proposed method was increased by more than 4.67% and 15% in accuracy and precision compared to other reported works.