Scholarlink

<p>Stage visual effects are essential for conveying emotion and maintaining narrative flow in theatrical performances, as lighting dynamics and visual transitions directly influence audience engagement. Traditional manual control limits adaptability and increases rehearsal effort, while existing automation lacks intelligent learning, causing slow convergence and unstable performance under dynamic conditions. This research proposes a Catfish Particle Swarm Optimized Double Deep Q-Network (CPSO-DDQN) for adaptive stage visual effect optimization. A Stage Visual Effects Motion Dataset containing 5006 synchronized multimodal samples with 3 audio and 3 video reference files, including rehearsal video, skeletal motion sequences, scene annotations, and music tempo signal, was constructed. Three dataset splits (60:40, 70:30, and 80:20) were used for evaluation. Preprocessing applied Histogram Equalization for visual enhancement, the Wiener Filter for audio denoising, and Z-score normalization for motion standardization. Feature extraction employed CNN models for visual and motion representations, while DWT captured discriminative audio features. Feature-level fusion integrated all modalities into a unified representation, improving correlation modeling. CPSO-DDQN combines swarm-based exploration with double Q-learning; CPSO maintains exploration diversity and prevents premature convergence, while DDQN captures reward distributions for robust policy optimization. Implemented in Python with TensorFlow and PyTorch. Across all splits, training accuracy ranged from 95.75 to 98.45%, while testing accuracy ranged from 92.60 to 97.20%. With recall ranging from 94.85 to 97.60% (training), and 91.80 to 96.40% (testing). Highest performance occurred with the 80:20 split. Overall, the CPSO–DDQN approach enables intelligent and adaptive stage visual effect control, reduces manual intervention, and improves the consistency and responsiveness of visual transitions during performances.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Adaptive stage visual effect optimization using CPSO–DDQN for dynamic theatrical environments

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年8月2日

查看全文
  • Chuan Fang,
  • Ruoxi Wang,
  • Dan Sun

摘要

Stage visual effects are essential for conveying emotion and maintaining narrative flow in theatrical performances, as lighting dynamics and visual transitions directly influence audience engagement. Traditional manual control limits adaptability and increases rehearsal effort, while existing automation lacks intelligent learning, causing slow convergence and unstable performance under dynamic conditions. This research proposes a Catfish Particle Swarm Optimized Double Deep Q-Network (CPSO-DDQN) for adaptive stage visual effect optimization. A Stage Visual Effects Motion Dataset containing 5006 synchronized multimodal samples with 3 audio and 3 video reference files, including rehearsal video, skeletal motion sequences, scene annotations, and music tempo signal, was constructed. Three dataset splits (60:40, 70:30, and 80:20) were used for evaluation. Preprocessing applied Histogram Equalization for visual enhancement, the Wiener Filter for audio denoising, and Z-score normalization for motion standardization. Feature extraction employed CNN models for visual and motion representations, while DWT captured discriminative audio features. Feature-level fusion integrated all modalities into a unified representation, improving correlation modeling. CPSO-DDQN combines swarm-based exploration with double Q-learning; CPSO maintains exploration diversity and prevents premature convergence, while DDQN captures reward distributions for robust policy optimization. Implemented in Python with TensorFlow and PyTorch. Across all splits, training accuracy ranged from 95.75 to 98.45%, while testing accuracy ranged from 92.60 to 97.20%. With recall ranging from 94.85 to 97.60% (training), and 91.80 to 96.40% (testing). Highest performance occurred with the 80:20 split. Overall, the CPSO–DDQN approach enables intelligent and adaptive stage visual effect control, reduces manual intervention, and improves the consistency and responsiveness of visual transitions during performances.