Tracking Aktivitas Manusia Dalam Ruangan Menggunakan Kalman Filter

Abstract

With the increase of many households using surveillance systems for security, unlike office and mall buildings which own security human resources, house owners make time to manually check security footages. This gives house owners inconvenience and owning a security system seems more tedious, therefore inefficient if the footage is rarely seen. AI makes it possible to automatically check security footages. Using footage taken from a special room in home, susceptible for security breach and strategic, for example living room. This App works as a tool for homeowners, giving information about indoor activity. Information is given in a form of video footage of human tracking results containing trajectory line and activity log. Hence home owners will be able to supervise certain rooms and human behavior. Adopting point tracking as a detecting method, the target object is detected using background subtraction and image preprocessing to obtain centroid point, an input for statistical prediction of Kalman filter. Testing results showed that RMSE of Kalman filter prediction is higher than background subtraction when compared to true location, therefore background subtraction is used for Kalman filter’s RMSE. Resulting in RMSE for two scenarios are 85,08 and 89,28, this app also shows overall accuracy of 65,43%, precision of 70,56% and recall of 63,18% in total.