Automatic Alert Generation against Pre-defined Rules-set for Perimetric Security of Sensitive Premises using YOLOv3

Autor: Muhammad Asif, Azeem Ilyas, Syed Muhammad Sajjad, Adnan Masood
Rok vydání: 2019
Předmět:
Zdroj: 2019 15th International Conference on Emerging Technologies (ICET).
Popis: Adaptation of artificial intelligence (AI) based solutions at sensitive locations is growing rapidly. Use of these techniques along with surveillance cameras has become a primary requirement of smart cities to convert data into intelligible information. Consequently, these solutions are minimizing the effort of training and reliance on human resource. In this research paper, we have devised use cases of object detection with focus on human and luggage detection in real-time using a Convolutional Neural Network (CNN) technique YOLOv3. This technique, trained on MS-COCO dataset, has not been able to produce desirable results when tested on images from subcontinent region containing luggage, human or both. As a case study, we enhanced MS-COCO dataset by incorporating our own collection of realistic images. The study is carried out on a commodity hardware to strengthen our claim to use technology over humans. The proposed solution is developed for analysing video streams in real time against a predefined rules-set. Idea to automate the process of surveillance at strategic locations without human intervention opens a new window of research for literary community to develop cost effective solutions.
Databáze: OpenAIRE