Lightweight CNN model: automated vehicle detection in aerial images

Efficient vehicle detection has played an important role in Intelligent Transportation Systems (ITS) in smart cities. With the development of the Convolutional Neural Network (CNN) for objection detection, new applications have been designed to enable on-road vehicle detection algorithms. Therefore,...

Full description

Saved in:
Bibliographic Details
Published in:Signal, image and video processing image and video processing, 2023-06, Vol.17 (4), p.1209-1217
Main Authors: Momin, Md Abdul, Junos, Mohamad Haniff, Mohd Khairuddin, Anis Salwa, Abu Talip, Mohamad Sofian
Format: Article
Language:English
Subjects:
Algorithms
Artificial neural networks
Boxes
Computer Imaging
Computer Science
Feature maps
Image Processing and Computer Vision
Intelligent transportation systems
Lightweight
Multimedia Information Systems
Object recognition
Original Paper
Pattern Recognition and Graphics
Signal,Image and Speech Processing
Vehicles
Vision
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Efficient vehicle detection has played an important role in Intelligent Transportation Systems (ITS) in smart cities. With the development of the Convolutional Neural Network (CNN) for objection detection, new applications have been designed to enable on-road vehicle detection algorithms. Therefore, this work aims to further improve the conventional CNN model for real-time detection on low-cost embedded hardware. In this study, a lightweight CNN model is proposed based on YOLOv4 Tiny to detect vehicles from the VEDAI dataset. In the proposed method, one additional scale feature map is added to make a total of three prediction boxes in the architecture. Then, the output image size of the second and third prediction boxes are upscaled in order to improve detection accuracy in detecting small size vehicles in the aerial images. The proposed model has been evaluated on NVIDIA Geforce 940MX GPU-based computer, Google Collab (TESLA K80) and Jetson Nano. Based on the experimental results, this study has demonstrated that the proposed model achieved better mean average precision (mAP) compared to the conventional YOLOv4 Tiny and previous works.
ISSN:1863-1703
1863-1711
DOI:10.1007/s11760-022-02328-7