Suspect Identification Based on Facial Recognition from CCTV Using Hybrid Bat Algorithm

The incidence of criminal activity is rising steadily, necessitating the use of surveillance cameras to monitor and control such activities effectively. However, traditional video evaluation by human guards and manual image processing often fails to accurately analyze footage, especially when individuals use disguises like glasses, masks, hats, or different hairstyles. To address this issue, this research proposes a surveillance system that leverages CCTV footage to detect and identify individuals attempting to conceal their identities. The system first detects facial areas using the Multi-task Cascaded Convolutional Networks (MTCNN) method. Once a facial area is detected, the system performs face identification by matching facial features of suspects with those captured on CCTV. If a match of 70% or higher is found, the individual is flagged as a suspect, with the output showing bounding boxes around the face labeled with their name. To enhance the system's efficiency, the FaceNet embedding model is integrated with the Hybrid Bat Algorithm (HBA) to reduce the dimensionality of face embeddings. HBA, which combines modifications from the Bat Algorithm (BA) with Differential Evolution (DE) techniques, effectively optimizes the process. Evaluation results showed the best performance with a maximum iteration of 50 and a population size of 30, achieving an accuracy of 0.877, recall of 0.830, precision of 0.907, and f1-score of 0.867. The face embedding dimension was reduced from 512 to 201 using HBA.