All Volumes/Issues

Volume 5, Issue 1

Title : A Modified Markov Random Field and CNN-Based Approach for Aerial Road Segmentation

Author : Prashant Kumar Bairwa-1, Suraj Yadav-2

Abstract :

Artificial neural network-based systems currently demonstrate the highest accuracy among various methods for road detection and segmentation. However, for real-world applications such as autonomous navigation and military operations, significantly higher precision is essential. To address this requirement, this study explores a hybrid approach that integrates an Optimized Markov Random Field (OMRF) model with parameter estimation from Convolutional Neural Networks (CNNs) for detecting and segmenting roads from aerial imagery. The hybrid framework enhances performance by leveraging both pixel-level and region-level information, achieved by factorizing the overall likelihood into pixel-wise and regional likelihood components. The output images are classified into road and non-road segments, and the effectiveness of the proposed system is evaluated using confusion matrix metrics. The combined OMRF-CNN model achieves an impressive accuracy exceeding 99.5% for road detection in airborne images.

Keywords: Image Processing, Detection of Road Region, Segmentation of Road region, CNN, ANN, Neural Network, Markov Random field

Download File

---