Indian Science Technology and Engineering facilities Map
Supplier Map
Service Map


Patent Details

Indian Institute of Technology (IIT) Patna 
System And Method For Detection Of Banned Objects From Images In Real-Time Using Intelligence At The Edge 
Dr. Rajiv Mishra, Mr. Himnshu Shekhar, Mr. Sourasekhar Banerjee, Mr. Yashwant Singh Patel 
Indian Institute of Technology (IIT) Patna 
Application type:
Applied - 2020-02-15 
Application Number:
Applied at:
Entered by:
Indian Institute of Technology (IIT) Patna 
The traditional way of detecting real-time objects using the cloud computing system was incurring substantial delay & cost in sending the raw data over the internet towards cloud for inferencing consumes high bandwidth, make the cloud busy, and it is costly due to public cloud usage for computation needs the intervention of the cloud on every action, hence not suitable for latency-sensitive and mission-critical operations. For latency-sensitive applications, the action must be taken close to the source( i.e., not at cloud). The proposed invention discloses a low cost distributed deep learning trained Cloud-IoT-Edge system to detect and classify banned objects in real-time with accuracy using low resource general-purpose computing hardware & devices made capable of inferencing for the response in realtime. The proposed invention solves the problem of latency sensitivity and massive bandwidth consumption using an integrated Cloud-IoT-Edge system. Here, Cloud- IoT-Edge System uses general-purpose computer hardware (as Edge System) connected to the camera ( i.e, IoT device) occasionally connected to Cloud for deriving/updating intelligence at the edge using dockers. Firstly, banned object detection model that was using faster R-CNN (Region- Convolutional neural network) was deployed on the edge device after dockerizing that object detection model, and it was capable of detecting the banned objects in real-time. If the detected object accuracy was not satisfiable(below some predefined threshold) or banned object classified wrongly which can be the case because object detection model does not include real-time data, then we sent the real-time data to the cloud services in our case Azure custom vision and there we made another object detection model using real-time images, performed training and testing there and after achieving the desired accuracy of the banned object we again do containerization of the model and deploy on the edge devices and then make prediction at the edge itself again and this process is repeated . The model should be retrained with new captured dataset after a particular time to improve the accuracy and lower the misclassification rate. 

I-Mitra(आई-मित्र) Welcomes You..
It has always been the basic tenet of the Government of India, in generously funding R&D efforts at academic institutions over the years, that facilities established through such support be made available to those needing them and qualified to make use of them for their own research work

However, this was never easy or straightforward for, among other reasons, there was no ready source of information of what facility was available and where. Thanks to the Web, it is much easier today to have a national and regional “inventory of resources”, so as to match users with the resources they need, and to do all this in an efficient and transparent manner.

This can lead to a leap in R&D productivity and greatly enhance the effectiveness of public investment. This is the motivation behind I-STEM.
read less <<
Visitor Hit Counter
Hosted at Indian Institute of Science
Copyright © 2020 I-STEM. All rights reserved.
Audited by: STQC Bengaluru.