Aerial detection of ground moving objects

Abstract

Automatic detection of ground moving objects (GMOs) from aerial camera platforms (ACPs) is essential in many video processing applications, both civilian and military. However, the extremely small size of GMOs and the continuous shaky motion of ACPs present challenges in detecting GMOs for traditional methods. In particular, existing detection methods fail to balance high detection accuracy and real-time performance. This thesis investigates the problem of GMOs detection from ACPs and overcoming the challenges and drawbacks that exist in traditional detection methods. The underlying assumption used in this thesis is based on principal component pursuits (PCP) in which the background of an aerial video is modelled as a low-rank matrix and the moving objects are modelled as sparse corrupting this video. The research in this thesis investigates the proposed problem in three directions: (1) handling the shaky motion in ACPs robustly with minimal computational cost, (2) improving the detection accuracy and radically lowering false detections via penalization term, and (3) extending PCP’s formulation to achieve adequate real-time performance. In this thesis, a series of novel algorithms are proposed to show the evolution of our research towards the development of KR-LNSP, a novel robust detection method which is characterized by high detection accuracy, low computational cost, adaptability to shaky motion in ACPs, and adequate real-time performance. Each of the proposed algorithms is intensively evaluated using different challenging datasets and compared with current state-of-the-art methods.