Prototype Facial Detection and Tracking System Using Digital Image Processing Assembled with A Mechanical Arm

Abstract

Face detection and tracking systems use digital image processing to analyse and identify human faces. This technology is widely applied in various fields such as security systems, personal tracking, and customer behaviour analysis. In the military, face tracking technology can be developed into an automatic targeting system by shifting the focus from general objects to human faces. When integrated with a gun turret control system, it becomes an efficient weapon. The system's operation starts with face detection using digital cameras to capture images, followed by image analysis software to locate and identify faces within frames and then track them. Once a face is detected, the centroid coordinates are sent to an Arduino robotic arm to track the face's movements within subsequent frames. The robotic arm adjusts its angle and position based on the information received from the system, targeting the human face. The experiment utilized four algorithms for face detection: the Viola-Jones algorithm, Dlib library, Convolutional Neural Networks (CNN), and Multi-task Cascaded Convolutional Networks (MTCNN). Performance measurement using a confusion matrix showed that the CNN algorithm had the highest efficiency, with a precision of 0.95, recall of 0.92, and F1 score of 0.93. Therefore, CNN is the most effective algorithm for face detection. The performance of the robotic arm combined with the face detection algorithms demonstrated that the CNN algorithm was the most accurate in identifying and tracking human faces, even in low light or complex environments. Testing the robotic arm's rotation at 45 and 135 degrees achieved an 82.5% success rate in face tracking.