Investigating the efficiency of deep learning based security system in a real-time environment using YOLOv5

Surveillance Systems Application based on deep learning algorithms is speedily growing in a broad range of fields such as Facial Recognition, Real Time Attendance Systems etc. Identifying several appearances in a real time environment is very crucial due to its difficult and heterogenous environmental conditions and blocking effects. We used state-of-the-art YOLOv5 model for investigating the efficiency of surveillance system with very limited experimental analysis. We used Face Detection Dataset & Benchmark (FDDB) and Celebrity Face Recognition (CFR) Dataset for training from scratch and for testing over YOLOv5 and private dataset taken from run-time video stream. Experimentations showing that we got 93% accuracy on FDDB on the other hand 99% accuracy on the tailored dataset. Comparison has been made for the analysis showing that our algorithm has produced better outcomes with the predecessor editions of YOLOv5 like YOLOv4 and YOLOv3 respectively. The aforementioned models are also validated over the run-time streaming, and it has the ability to recognize many faces with maximal precision.