An adaptive wavelet transform application to multiple targets tracking in the air
Rui Wang, Jingwen Xu, Yan Fei Zhao · 2017
Motion estimation (ME) has been extensively applied in the computer vision, including vision-based target tracking in the air. For attaining robust performance in noise and handling obscuration or the parallel moving nearby target interferences trajectories, the ME algorithm based on the spatio-temporal continuous wavelet transform (CWT) with a band pass velocity Kalman Filter in the transform domain is proposed, which is designed to conduct the efficient multiple target tracking in an adjustable spatio-temporal processing block. The CWT allowing for the definition of three energy densities over a subset of the CWT parameter space, which accompanied with Kalman Filter has been employed to capture motion information over multiple frames and proved excellent in velocity selectivity. To best handle interferences among multiple nearby targets, a simpler and robust solution called as self-adaptive rotation of coordinates and a practical functional relation between the target radius, speed and the scale parameter is developed. The presented novel joint processing technique using expectation-maximization based Gaussian mixture estimation, together with a global nearest neighborhood algorithm to perform data association, achieves maintaining kinematic trajectory of every targets either in linear or nonlinear motion. Examples with synthetic data and real data taken in air surveillance are given to demonstrate the effectiveness of these proposed strategies.