Detection of Moving Objects for Aerial Surveillance of Arbitrary Terrain
Zusammenfassung
The detection of moving objects in aerial video sequences is a common application in safety and environmental monitoring. The challenge is the non-static camera, which is moving together with an aerial vehicle. To detect local changes due to movement of ground objects in such a scenario, the displacements of image pixels resulting from the motion of the camera need to be compensated. The most common method is to use a projective transformation and assume the observed scene to be planar. However, this is only valid for very high altitudes. It fails otherwise and results in falsely detected local motion. This work addresses the problem in two ways. After analyzing the error resulting from motion parallax, two detectors for moving objects in non-planar scenes are presented. One is based on a motion parallax model and one on a smooth optical flow approach. Following this, a motion compensation method for non-planar scenes is presented, allowing the use of image differences based methods ...
Schlagworte
- I–XII
- 1–9 1 Introduction 1–9
- 1.1 Motion Estimation and Parameter Extraction
- 1.2 Problems of Current Ground Motion Models
- 1.3 Contributions
- 1.4 Outline
- 10–33 2 Basic Principles 10–33
- 2.1 SceneModel
- 2.2 Camera Model
- 2.3 Epipolar Geometry
- 2.4 Essential and Fundamental Matrix
- 2.5 Projective Transformation and the Homography
- 2.6 Moving Object Detection by Background Subtraction
- 2.7 Motion Estimation from Image Sequences
- 2.8 Dense Optical Flow
- 34–57 3 Planar Landscape Error Model 34–57
- 3.1 Aerial Surveillance Scene Model
- 3.2 Vertical Aerial Photo
- 3.3 Motion Parallax Displacement
- 3.4 Camera with a Tilt Angle
- 3.5 Arbitrary Camera Orientation
- 3.6 Height Restrictions based Outlier Detection
- 3.7 Detectability in Dependence of Speed and Direction of Motion
- 58–72 4 Multi-Planar Landscape Model based on Triangle Meshes 58–72
- 4.1 Mesh Creation
- 4.2 Outlier Removal and Moving Object Detection
- 4.3 Mesh-based Motion Compensation
- 4.4 Accuracy Analysis of the Mesh-based Approach
- 73–92 5 Experiments 73–92
- 5.1 Motion Vector Classification
- 5.2 Motion Compensation Performance
- 5.3 Moving Object Detection Performance
- 93–96 6 Summary and Conclusions 93–96
- 97–106 Bibliography 97–106
