ALGORITHMS INTEGRATION OF OBJECT DETECTION, FACE RECOGNITION, AND ABNORMAL BEHAVIOR USING STATE MACHINE

• Main Author: Fauzi Nuryasin, Muhammad
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/73242
• Institution: Institut Teknologi Bandung

Artificial intelligence research in image processing has been rapidly advancing recently. Image processing can be utilized in surveillance camera systems. Surveillance camera systems rely on humans to interpret the captured images in video recordings. However, manual observation by humans is susceptible to distractions and fatigue. Therefore, this research designs three modules with different goals and functions, integrated into one system to assist human supervision. The three modules consist of object detection, face recognition, and anomaly behavior detection, integrated with state machine. This research utilizes the Histogram Oriented Gradient-Support Vector Machine (HOG-SVM) method for object detection, the utilization of deep learning Convolutional Neural Network (CNN) with an adaptation of transfer learning method combined with Visual Geometry Group-16 (VGG16) architecture for face recognition, and a spatiotemporal autoencoder based on spatial and temporal dimensions relying on loss values and thresholds to identify anomaly behavior. Standard metrics such as accuracy, precision, recall, and F1-score are used to test the three models of these modules. The three modules are integrated with a state machine to form a unified system. The overall system can recognize and differentiate between normal and abnormal conditions that occur. The overall system accuracy reaches 84%, precision value is 84.615%, recall value is 94.286%, and F1-score is 89.189%.