Optimized Sensor-Based Interaction System for Digital Media Games
Yu Ma · Journal of Engineering Project and Production Management · 2026
As technology advances at an accelerating pace, there is an increasing demand for games using digital media that offers advanced social interactions and an enhanced user experience. However, during 3D game development, issues such as inadequate capture of human body posture information continue to be a challenge. To address this, the study introduces a training method for a digital media game interaction system based on the dual-sensor Levenberg-Marquart (LM) algorithm to improve the system’s computational capabilities. This approach also employs Composite Fields for Human Post Estimation (PIFPAF) to extract key features of human body movement points, boosting the system’s ability to handle complex scenarios. Experimental results indicate that when the algorithm iterates between 200 and 250 times, the player’s interaction loss rate ranges from 0.03% to 0.04%. The proposed method responds to all four types in under 0.250 seconds. In human-computer interaction, delays must remain below the human perception threshold to maintain immersion. Therefore, the system’s ability to respond rapidly ensures that game interactions occur closer to real-time while effectively avoiding operational lag and guaranteeing a more natural and smooth interaction for its users. In contrast, the bidirectional long short-term memory training method is 0.350 seconds, and its perceived delay is more likely to disrupt the player’s immersive state. These outcomes demonstrate that the proposed training method effectively captures human body information and facilitates Human-Computer Interaction feedback. This research contributes to the future development of 3D digital media games by supporting more advanced interaction commands.