Active object detection and tracking are important parts of intelligent unmanned aerial vehicles, which have drawn extensive attention from numerous researchers. However, many problems still remain open. In this thesis, we lucubrate on geographical feature extraction, moving object detection and target tracking techniques from aerial images to meet the requirement for active UAVs. The main contributions are summarized as follows: (1)Based on curvilinear characteristics of roads in low resolution aerial images and kernel-based density estimation analysis, we propose a novel line extraction and delineation approach by using LWF (Local Weighted Features). Experimental results demonstrate that our algorithm is very fast, usually need not tune parameters, is robust against noises and has high environmental applicability. (2)A new method for road detection from aerial images based on total variations and mathematical morphology is proposed.Strict experiments show that this algorithm is efficient and has high stability and adaptability. (3)We propose an algorithm for ground moving object detection based on spatio-temporal analysis. Experimental results indicate that this algorithm has good performances and high environmental applicability. Additionally, our method is very efficient and can meet the requirement for real-time applications. (4)In order to copy with tasks such as vehicle relocation when tracking fails or new vehicle detection during the tracking procedure in high resolution aerial images, we propose a novel on-road vehicle detection algorithm based on texture and color analysis. Experiments reveal the efficiency and effectiveness of our method. (5)The classical MS tracker is improved to get a novel robust tracker with global tracking ability which is termed as APMS (Adaptive Pyramid MS). Evaluation results on a number of videos and its application to a tracking and pointing subsystem demonstrate that our method is robust against initialization errors and can copy with situations such as camera vibration and partial occlusions.
修改评论