In this paper, a deep learning-based facial recognition system is outlined that specifically attempts to handle challenges typically encountered in real-world scenarios, like low light and blurry images, and partial occlusion. Designed for real-time applications in various conditions, such as blurred or night images, we used CLAHE (Contrast Limited Adaptive Histogram Equalization) and gamma correction as image enhancements to significantly increase the visibility of facial features. We also applied dlib-based facial landmark detection for precise localized refinement of features. The key innovation of our system is its dynamic face registration; hence, new faces can be easily included in real-time environments without updating the database manually. Therefore, extensive experiments have been performed to demonstrate how the system can maintain high recognition accuracy under quite diverse and unpredictable conditions. This outperforms traditional methods even in challenging environments. It is highly suitable for applications in security, surveillance, and personalized user interactions for this reason. Future work will include the combination of super-resolution techniques and multimodal biometric systems that, in turn, will further enhance the performance of the recognition system, making it more applicable to a wider use case.