Kyungjune Lee

Finding the preferred camera trajectory is an important sequence for making three-dimensional (3-D) human music videos, which is a very challenging task. To achieve this, several studies have attempted to combine human visual information with 3-D video camera trajectories. However, previous studies have been limited to specific situations, and their applications have been limited to certain scenarios. To address this limitation, we present a novel method for camera trajectory planning. Toward this goal, we first developed a perceptual metric to quantify satisfaction with the trajectory of a camera capturing a moving 3-D human model. We introduce new metrics designed to derive three features: 1) the normalized limb length sum incorporating part attention, 2) the normalized human gaze vector, and 3) the normalized stability viewpoint. Along with these features, we inherit a commonly used feature: 4) normalized projected area. By maximizing these four features, we obtain an intermediate camera trajectory. Furthermore, we introduce new metrics to obtain two features that potentially disrupt visual immersion: 1) boredom resulting from repetitive or stationary trajectories, and 2) visual discomfort caused by promiscuous camera movements.With intermediate camera trajectory, we can find an optimal camera trajectory by minimizing these two adverse features. To encapsulate these metrics, we propose a novel camera trajectory planning system (CTPS). By conducting rigorous experiments, we demonstrate that a camera trajectory from the proposed algorithm is preferred and exhibits a 22.1% higher similarity to the camera trajectory of professionals compared to existing solutions.