Surveillance systems play a vital role in the safety and protection of people, although they are often faced with problems such as inadequate lighting, occlusions, and pose variations. Conventional face recognition methods typically fail in this kind of setting and deep learning algorithms require significant computational resources, which is incompatible with real time use on a compact IoT device. We present a lightweight texture-feature-based smart surveillance framework in this paper, combining Max-Regional Fuzzy Encoding (MRFE) and the Gray-Level Co-occurrence Matrix (GLCM) statistics and the variations of Local Ternary Pattern (LTP). The MRFE-GLCM component is complementary to enhance strength, as it combines fuzzy regional encoding with global texture encoding, whereas LTP variants are resilient to illumination and noise variations. These characteristics are merged into a small form and categorized with the help of SVM, KNN, and Decision Tree classifiers. It is experimented on benchmark datasets (ORL, YALE, Extended Yale B) that the proposed hybrid has better accuracy than traditional LBP variants and CNN-LBP hybrids, with low computational complexity. Based on a light weight structure, the framework can be used in real time smart surveillance applications, such as secure access control, smart monitoring, and IoT-based public safety systems. The proposed framework should be computationally lean and efficient. Extended performance analysis confirms the robustness and suitability of the model for real-time IoT-based surveillance.