Author(s): Xin Liu; Jinghong Deng; Yi Xiao
Linked Author(s):
Keywords: Kalman filter; Hungarian algorithm; Particle recognition; Bedload particle movement
Abstract: Addressing the inadequacy of traditional image tracking algorithms in accurately identifying sediment particle motion characteristics under complex dynamic conditions, this paper presents an enhanced trajectory tracking algorithm for bedload particles based on the integration of Kalman filtering and the Hungarian algorithm. The proposed method initiates with Gaussian Mixture Differential modeling to precisely differentiate image backgrounds from particle motion regions. Subsequently, a dual noise reduction strategy employing median filtering and morphological opening operations is applied to eliminate noise induced by uneven illumination and adherence of motion regions due to particle collisions. Thereafter, Kalman filtering is introduced for optimal estimation of movement positions, which is then coupled with the Hungarian algorithm to match the varying frame positions of the same particle, thereby yielding precise multi-particle trajectories. Experimental outcomes have demonstrated that the improved algorithm significantly enhances the recognition accuracy of particle coordinates and motion trajectories across varying flow intensities. Specifically, the average relative error in identified particle coordinates has been reduced by 70% compared to conventional methods, while the coefficient of dispersion of coordinates in sequential images has decreased by 66%. Traditional approaches, in contrast, suffer from a 30% miss-detection rate and incorrectly identify noise as particle trajectories. The proposed improvement effectively circumvents these issues of missed and false identifications, achieving accurate tracking of all particle motion paths without such discrepancies.
Year: 2024