Abstract:In this paper we first contribute a large scale online study (N=400) to better understand aesthetic perception of aerial video. The results indicate that it is paramount to optimize smoothness of trajectories across all keyframes. However, for experts timing control remains an essential tool. Satisfying this dual goal is technically challenging because it requires giving up desirable properties in the optimization formulation. Second, informed by this study we propose a method that optimizes positional and temporal reference fit jointly. This allows to generate globally smooth trajectories, while retaining user control over reference timings. The formulation is posed as a variable, infinite horizon, contour-following algorithm. Finally, a comparative lab study indicates that our optimization scheme outperforms the state-of-the-art in terms of perceived usability and preference of resulting videos. For novices our method produces smoother and better looking results and also experts benefit from generated timings.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is how to generate more visually appealing camera motion trajectories for quadrotor drones. Specifically, the authors focus on:
1. **Difficulty for users to control spatio - temporal distances**: When specifying key frames, users often find it difficult to accurately consider spatio - temporal distances. This can lead to the generated camera motion trajectories being too fast or too slow in certain segments, affecting the aesthetics of the video.
2. **Trade - off between optimizing global smoothness and maintaining time control**: For novice users, optimizing global smoothness can significantly improve the quality of the video; for expert users, they need precise time control to achieve specific composition effects. Finding a balance between these two is a challenge.
3. **Limitations of existing methods**: Existing quadrotor drone camera tools usually fix the time of key frames as hard constraints, which limits the ability to generate smooth motion. In addition, these methods may cause excessive changes in camera speed when dealing with complex scenes, resulting in unpleasant visual effects.
To solve these problems, the authors propose a new optimization method that can optimize the fitting of position and time references while maintaining the user's control over the reference time, thereby generating globally smooth trajectories. Specifically, their contributions include:
- **Large - scale online study**: Through a large - scale online study (N≈400), the authors have a better understanding of people's aesthetic perception of aerial videos and find that global smoothness is crucial for video quality.
- **New optimization algorithm**: A variable, infinite - field - of - view contour - following algorithm is proposed, which can re - allocate the robot position and camera angle during the optimization process, while treating the key - frame time and position as well as the reference speed as soft constraints. This allows users to trade off between path - following fidelity and time fidelity.
- **Experimental verification**: Through laboratory studies, the authors compare videos generated by different methods, and the results show that their method is superior to existing methods in terms of usability and video quality, both for novice and expert users.
In summary, this paper aims to optimize the motion trajectories of quadrotor drone cameras to make them more visually appealing and find a reasonable balance between novice and expert users.