False Alarm Filtering within Camera Surveillance using an External Object Classification Service
Jonathan Lundholm, Paul Maxwell Steneram Bibby · 2017
Axis cameras can detect motion and track object, but lacks object classification. This often gives rise to false alarms, which are triggered by non-human activity. The cameras cannot tell the difference between a tree blowing in the wind, which is a non-human activity, and a person breaking an entry, which is a human activity. This becomes a problem as false alarms can be resource consuming for operators. Especially during windy weather, when there are lots of alarms and it is hard to know which a genuine, and which are not. This thesis has looked at the viability of using an external classification service for identifying objects, and using it to filter false alarms by distinguishing human activity from non-human activity. The classification service has been implemented as a distributed system deployed in the cloud, and the thesis has used footage and videos provided by Axis to test the viability of the service. It has also looked at ways to reduce the network traffic by manipulating the image's size and compression level. Results show that the service has great potential in classifying objects and filtering false alarms. However it lacks depth into algorithms for the selection of images to use when classifying an object. The service shows great promise, but needs further improvements and investigation to properly determine the viability of being used in the field.