An approach to Instance segmentation of Quadcopters for Suspicious Surveillance

The development of an aerial surveillance system could be very effective in meeting surveillance system challenges as well as for upholding the safety and security of the people. A cost-effective and efficient solution is Quadcopters, which are used for aerial surveillance are outfitted with cutting-edge image processing technology that captures detailed photographs of every object beneath. A quadcopter-based solution is proposed to monitor desired locations for any unusual activity of quadcopters, such as the desired surveillance. In this paper, an approach to Instance segmentation on the different Detectron based Deep learning models for generating an alert for security personnel after detecting any unusual activity is proposed. The Deep learning models relies on detecting the anomalies in received image or video streams captured. In the initial feature extraction step, a well-known Faster RCNN algorithm is modified for fast learning with feature reduction. The ability of five different types of CNNs to identify objects of interest in surveillance images was tested. The fastest ResNet-50 based Faster RCNN with the highest average precision, performed admirably in threat detection and average precision with IOU and accuracy of the system across all categories.