Methods and apparatus for implementing and using techniques for constructing an artificial neural network for a particular surveillance situation are provided, including computer programs. A certain number of object classes are selected that represent characteristics of the monitoring situation. These object classes form a subset of the total number of object classes in which the artificial neural network is trained. A database containing activation frequency values for neurons in an artificial neural network is accessed. The activation frequency values are values depending on the object class. Those neurons with activation frequency values less than the threshold for a subset of the selected object classes are removed from the artificial neural network.
展开▼
