Research on Film Advertisement Interaction Control Method Combining Deep Generative Models and Gesture Recognition

This paper proposes an interaction control method that integrates gesture recognition with conditional generative models to enhance the personalized responsiveness and user engagement of film advertisement systems. Traditional advertisement playback methods generally lack interactivity and are unable to dynamically adapt to user behavior. To address this issue, an end-to-end generative framework is constructed, mapping gesture recognition results into controllable generation signals, thereby enabling real-time linkage between user natural gestures and advertisement content. The system utilizes a Transformer-based dynamic gesture recognition model to extract semantic embeddings, which are then fed into a conditional generative adversarial network (cGAN) to produce advertisement keyframes aligned with user intent. A dynamic response-content synchronization mechanism is further incorporated to optimize system latency and responsiveness. Experimental results demonstrate that the proposed method outperforms conventional advertisement control systems in several critical aspects, including recognition accuracy, content generation quality, response speed, and user satisfaction. The average system response time is maintained under 200 milliseconds, and user satisfaction scores reach 4.6 out of 5, validating the system's significant improvement in interactivity and usability. The proposed system not only offers a unified framework for advertisement content generation and interaction control but also provides a new technical paradigm for the design and implementation of multimodal human-computer interaction systems.