Hanh Tran

Research interests

I focus mainly on anomaly detection in video. A model of normality is trained from videos contain only normal events and then is used for anomaly detection and localisation in testing video frames. The testing videos have both normal and abnormal events.

In particular, I use deep convolutional auto-encoders. Auto-encoders are used in two different ways: (1) to learn feature representations which are used for learning model of normality and for detecting/localising anomaly. In this case, I also use one-class Support Vector Machine for model learning and anomaly detection/localisation; (2) to encode normal events and then decode them with a small reconstruction error.      

Research groups and institutes

  • Artificial Intelligence