This paper presents a low-cost head movement-based adaptive user interface (UI) solution for quadriplegic people to access devices, implemented with computer vision and an adaptive learning algorithm using Recurrent Neural Networks with Long Short-Term Memory (RNN-LSTM) to enhance accuracy and personalization. This system uses OpenCV for facial landmark tracking from visual input to classify head movements in real-time. The adaptive model provides a low-latency assistive technological solution and uses feedback to improve accuracy. Simulation results demonstrate that our proposed method, Vision-based with Adaptive Learning, outperforms existing methods in terms of accuracy, having a better accuracy of 92%. The proposed system minimizes the input latency averaging at 50 ms, with an improved adaptation rate and classification reliability, making it a highly effective solution for applications requiring precise and responsive gesture recognition.