Dharmender Saini, Narina Thakur, Rachna Jain, Preeti Nagrath, D Jude Hemanth, Nitika Sharma
Abstract
Unmanned aerial vehicles (UAV) technology has revolutionized the field globally in today’s scenario. The UAV technologies enabled the activities to be efficiently monitored, identified and analyzed. The principal constraints of the present surveillance system, along with closed-circuit television (CCTV) cameras, are limited surveillance coverage area and high latency in object detection. Deep learning embedded with UAVs has found to be effective in the tracking and monitoring of objects, thus overcoming the constraints mentioned above. Dynamic surveillance systems in the current scenario seek for high-speed streaming and object detection in real-time visual data has become a challenge over a reasonable time delay. The paper draws a comprehensive analysis on object detection deep learning architectures by classifying the research based on architecture, techniques, applications, and data-sets. It has been found that RetinaNet is highly accurate while YOLOv3 is fast.
Keywords Object detection, Convolutional Neural Network, surveillance
**View Full Paper in the Springer Source