WSEAS Transactions on Signal Processing
Print ISSN: 1790-5052, E-ISSN: 2224-3488
Volume 13, 2017
Real-time UAV Target Tracking System Based on Optical Flow and Particle Filter Integration
Authors: , , ,
Abstract: This paper presents a design and implementation of a real-time, vision-based target tracking system for unmanned aerial vehicle (UAV). The particle filter framework integrated with Lucas-Kanade optical flow technique to predict and correct the state of the moving target based on its dynamic and observation models. The optical flow estimates the corresponding feature points in the new image frame related to the previously detected/estimated points. The Maximum Likelihood Estimation SAmple Consensus (MLESAC) method is applied to estimate the ego-motion transformation matrix using the old and new sets of the feature points. This matrix is incorporated with the target dynamic model to give more accurate prediction results of its state. Two optimized types of features are extracted to build the target observation model. They include extended Haar-like rectangles and edge orientation histogram (EOH) features. A Gentle AdaBoost classifier is applied on these features to distinguish and choose the best predefined number of features that highly represent the target. The vectorization approach is used to reduce the calculation cost due to the matrix manipulations. The proposed tracking system is tested on different scenarios of the on-time modified VIVID database and achieved real time tracking speed with 95% successful tracking rate.
Search Articles
Keywords: Machine Vision, Image Analysis, Video Tracking, UAV Tracking, Lucas-Kanade Optical Flow, Bayesian Particle Filter, Ego-motion
Pages: 172-181
WSEAS Transactions on Signal Processing, ISSN / E-ISSN: 1790-5052 / 2224-3488, Volume 13, 2017, Art. #19