Matching of Multiple Moving Ground Targets Observed by Aerial Cameras Based on 2D Kinematics

This paper investigates the effectiveness of a proposed 3-layer, motion-based algorithm to match and re-identify detected ground targets across multiple aerial cameras. The main advantage of this method over traditional image-based matching and re-identification techniques is that there is no prior image-based training required; instead, simpler graph matching algorithms are used. The algorithm was proposed in an earlier study for ground-based detection of aerial targets. In this paper, it was applied to more general air-to-ground scenarios, where the cameras are non-parallel but pointing at the same 2D ground plane, and more than two targets are tracked at any point in time. Field tests demonstrated that the proposed algorithm is suitable for detecting and matching ground targets with long tracking history and distinct motions. However, its effectiveness is limited by occlusion effects, and when all targets are moving with similar heading and velocity. This work extends the application of the proposed algorithm to match and re-identify multiple ground targets under the scenarios described above.