Towards autonomous surveillance in real world environments

We present a portable system that is capable of providing autonomous surveillance in real-world environments. We aim to expand the functionality of surveillance systems by combining autonomous object recognition along with depth perception to identify the object and its distance from the camera. Such capability would prove invaluable to autonomous surveillance applications, where persons carrying any forbidden and/or dangerous objects are detected in real-time and appropriate warnings are signaled. We have selected Microsoft Kinect V2 system which includes built-in hardware implementation of algorithms to identify humzans in a complex real-world setting. In addition, the system can simultaneously track 6 people at any time and provide their skeletal joint diagrams for motion tracking. The current work deals with using the skeletal joint diagrams and depth maps to create a focus around the hand area of the people. Our developed algorithm deals with object detection after the segmentation of hands. We use machine learning techniques with establishment of training datasets that include the library of objects that we aim to detect. Finally, the complete signal processing software is implemented within a single board computer.