Object recognition and classification for security surveillance system using single board computer

This project presents a development of object recognition and classification for security surveillance system using single board computer, the system is able to recognize object type. Then, close circuit television (CCTV) systems may operate continuously or only as required to monitor a particular event. If something happen, person in charge will needs to access recorded video to see the situation, then CCTV also cannot detect and recognize any object. The project is to implement an object recognition method with Deep learning to detect object and classify. The report also describes the project to implement surveillance system for real time object detection to reduce monitoring activities by human to classify an object. By using recognition method, the system can recognize and classify any object that come through camera view and classification features will extract the object from recognition to classify type of object that comes through the camera. The project was developed using deep neural network method and SSD Mobilenet for detection. MobileNets is a neural network that is base network provide high level features for classification and SSD use of last convolutional layer on base networks for detection task. By using MS-COCO dataset that has 80 class of object helps this project to detect basic object which are human, animals and vehicle. This project uses a laptop and Raspberry Pi for comparison of accuracy. The data analysis takes accuracy data in different time and environment to test accuracy percentage and classification. For each test, 6 sample collected for analysis purposes. Afterwards, performance comparison between laptop and Raspberry Pi are made to compare the difference between in accuracy and classification performance. The result show that laptop accuracy is better which averaged of at 94.90% for laptop and 83.22% for Raspberry Pi. Raspberry Pi performance is found to be acceptable for surveillance system implementation as it smalls and requires less power.