This paper is published in Volume-6, Issue-6, 2020
Area
Computer Science Engineering
Author
Jithin P. Sajeevan, Prajval P., Shourya Mishra, Shwetha G. P.
Org/Univ
LRDC, Bangalore, Karnataka, India
Pub. Date
23 December, 2020
Paper ID
V6I6-1241
Publisher
Keywords
Deep Neural Networks (DNN), Region-Based Convolutional Neural Network (RCNN), You Only Look Once(YOLOv3), Single Shot Detection (SSD), Faster-RCNN with VGG-16

Citationsacebook

IEEE
Jithin P. Sajeevan, Prajval P., Shourya Mishra, Shwetha G. P.. DNN ‌Design‌ ‌for‌ ‌Object‌ ‌Detection in ‌Airport‌ ‌Runway‌ ‌Operations‌, International Journal of Advance Research, Ideas and Innovations in Technology, www.IJARIIT.com.

APA
Jithin P. Sajeevan, Prajval P., Shourya Mishra, Shwetha G. P. (2020). DNN ‌Design‌ ‌for‌ ‌Object‌ ‌Detection in ‌Airport‌ ‌Runway‌ ‌Operations‌. International Journal of Advance Research, Ideas and Innovations in Technology, 6(6) www.IJARIIT.com.

MLA
Jithin P. Sajeevan, Prajval P., Shourya Mishra, Shwetha G. P.. "DNN ‌Design‌ ‌for‌ ‌Object‌ ‌Detection in ‌Airport‌ ‌Runway‌ ‌Operations‌." International Journal of Advance Research, Ideas and Innovations in Technology 6.6 (2020). www.IJARIIT.com.

Abstract

In this project, we attempted to solve the problem of object detection in airport runway environments by leveraging DNNs. The DNN model developed will be able to classify objects and also accurately localize objects of different classes. The major aspect here is the task of object detection is going to be considered as a regression problem to object bounding box masks. It would be a multi-scale inference procedure that can generate high-resolution object detection at a minimal cost. Here we were able to test our custom dataset on three object detection models. The models considered were (1) Faster-RCNN with VGG-16, (2) YOLOv3 with darknet53, and (3) SSD Inception V2(coco) from the TensorFlow object detection API. All the above algorithms were trained using the Tensorflow framework. The paper gives a brief comparison in the performance of the above-mentioned algorithms when trained and tested while keeping the main goal to be object detection in airport runways. We trained the YOLOv3 with darknet53 on our custom dataset and were able to obtain the classifier accuracy for the bounding boxes to be 74.00% which was the best of the three algorithms and has outperformed many previous works. The YOLOv3 could be considered a fast training algorithm when the model is trained on a powerful GPU and could play a major role in the field in terms of real-time object detection scenarios.
Paper PDF