Date of Award


Document Type


Degree Name

Master of Science in Computer Science


School of Computer Science and Engineering

First Reader/Committee Chair

Jin, Jennifer


The detection of suspicious human activity is a crucial aspect of ensuring public safety and security. The aim is to identify suspicious behavior. To accomplish this, we employ the LRCN, a long-term recurrent convolutional network, to detect anomalous activity. It is important to consider the temporal data of the video when classifying suspicious behavior, and the framework uses a combination of CNNs and RNNs to analyze video frames and extract relevant features. The key milestones of this project include conducting research, collecting and pre-processing data, designing and training the model, and evaluating its performance. The resulting detection system can accurately identify suspicious behavior in real-time. To build the model, we used the KTH dataset, which includes 600 frames of walking and running, as well as the Kaggle dataset, which consists of 100 training videos. Our model analysis shows that the system and video can detect suspicious events with an accuracy of 86%, and we anticipate that this accuracy will improve as the dataset size increases.