This study explores enhancing drone capabilities in low-light environments using gesture and body pose recognition. Deploying a Crazyflie 2.1 drone integrated with sensors and AI modules, we leverage Google’s MediaPipe for real-time gesture detection and pose estimation under different lighting conditions. Compared to other traditional image-based methods, our approach with MediaPipe’s landmark-based models highlighted significant improvements in accuracy and computational efficiency. Results demonstrate that the landmark-based models not only perform robustly across different lighting conditions but also reduce dependency on high-quality images, making them suitable for real-time applications in challenging visual conditions. The use of these models in the drone’s autonomous navigation system shows potential improvements in how it operates and extends their use in important areas like search and rescue, military surveillance, disaster monitoring, and interactive technologies. This paper proves that landmark-based methods are better than traditional image classification, paving the way for more research into smart, responsive unmanned aerial vehicles for important uses including defense and safety.