Tracking software for owner of unattended object from video streaming

Abstract

This thesis proposes a design of smart surveillance software that candetect unattended objects and track the owner of the object

upon the successfulimplementation, this can decrease the workload of security staff. The technique is implemented by using image processing techniques along with machine learning. In the event of an unattended object being detected, the system alerts the security staff and displays tracking the owner of the object. The system consists of seven steps: 1.) ImageAcquisition, 2.) Image Detection, 3.) Event Recognition, 4.) Tracking, 5.) Ownerprobability, 6.) Configuration, and 7.) Evaluation. The experiments are conducted in order to measure the accuracy of each process in the system. This system uses twotypes of datasets which are PETS 2006 and self-generated datasets. The results show the accuracy of image detection process is 93.1%

the accuracy of the event recognition process is 96.1%

the accuracy of the tracking process is 89.33%

the overall accuracy of tracking in the system is 83.35%, and the overall accuracy of the system is 78.57%.