Assisted child-minding based on visual activity monitoring in a home camera surveillance system

Abstract

Taking care of a child can be a very demanding task nowadays. Accidents can occur if the child minder is not paying enough attention. Children might end up being injured or breaking stuff around the house. To help cope with this problem, a system that helps assist the child minder in monitoring the child is being proposed.

This project is mainly about developing a semi-automated system that obtains realtime surveillance footage from a home equipped with surveillance cameras. Using this footage, the user is then able to track a target (child) and monitor its movements. Also, the user is able to specify certain regions of the footage to be offlimits to the target. Whenever the target comes into these regions, the system would then notify the user so that he/she can take appropriate actions.

The system is developed using C++ and OpenCV library. First, the CAM Shift and Mean Shift algorithm were implemented to test how well both functions is able to track the child given in the dataset. Later on, Background Subtraction technique was then used to combine with Mean Shift for a better performing tracker. A function to specify Restricted Zone in the footage and an algorithm to sense whether the tracking target enters this zone is developed later on.

When first implementing CAM Shift and Mean Shift tracking algorithms, the program is unable to track the target as a whole throughout the sequence. However, after combining Mean Shift with Background Subtraction, the tracker’s performance improves. Also, using the Restricted Zone function, the user is notified by the system whenever the target enters a Restricted Zone specified by the user.

At the end of this project, the system is able to track the child inside the surveillance footage and monitor their movements. In future, as surveillance camera becomes more popular, developments of similar systems to aid man is evitable.