Since 2008, 8 persons have drowned in Limfjorden, Aalborg. To address the problem Aalborg Kommune would like to detect these incidents using Thermal Cameras. The origin of this master’s thesis is to carry out initial analysis on how to solve the problem using computer vision techniques. Three thermal cameras have been installed at the harbor. In this thesis an automatic surveillance system have been analyzed, developed and tested. From the thermal camera feed a person's position is automatically extracted and tracked thought the scene using a Kalman filter. To prevent fatal accidents, the system is able to detect if a person falls into the water using a trip-wire and optical flow. The fall detector algorithm is able to detect 100\% of all falls. The system only provides 0.08 false positive alarms per hour. Besides the fall detector a fall predictor has also been developed which will give an operator a warning before a person actually falls into the water. Test showed that the system is able to predict 23.67 \% of the trip-wire activations beforehand. To develop the system a dataset of 155 hours of thermal video have been recorded during night hours. Persons trajectories have been annotated for 56 hours to use for training and module test -- the remaining 99 hours have been used as an acceptance test.