Intruder Detection Using Face and Object Recognition

This work proposes an intelligent face recognition scheme for real-time monitoring on the basis of CCTV feeds, incorporating dynamic alerting and incremental learning for adaptive performance. The system derives facial features from frames, classifies unfamiliar vs. familiar individuals, and solicits users for manual confirmation of unrecognized faces. Once confirmed, new identities are inserted through online incremental learning so that the system continuously improves. Experimental results on benchmark datasets (LFW, IJB-A) and real-world CCTV footages prove high recognition accuracy (up to 97.2%) and low false acceptance/rejection rates. In addition, the system proves to be dependable in terms of real-time performance with alert latencies less than 2 seconds and continuous learning of new identities with 95.4% post-update accuracy. The advocated methodology provides an adaptive, scalable, and accurate solution for contemporary surveillance environments involving ongoing identity management.