부산대학교 도서관

During the last decade, human activity detection is increasingly attracting the attention of researchers, due to its numerous applications, such as in smart and automated shopping malls, hospitals, etc. Particularly, detecting human activity has been a challenge for researchers because complex situations may arise such as background clutter or changing illumination. To solve this issue, video segment classification cannot be tackled just as object identification. Therefore, it becomes inevitable to employ sequence-based techniques for video classification. In this paper, a Convolution Neural Network (CNN) is used in conjunction with Long Short-Term Memory (LSTM) to accomplish real-time human activity detection. In the proposed method, CNN serves as a spatial information detection algorithm from video while LSTM helps in the sequential tracking of identified objects quickly and accurately. This CNN-LSTM approach reduces the complexity of the model while also enhancing its accuracy along with enabling its real-time execution. Finally, a Raspberry Pi that functions as a standalone system is utilized for the implementation of the proposed CNN-LSTM approach. The results are presented and analyzed to solidify that the proposed standalone system can detect and classify events for real-time surveillance.