CNN based System for Road, Vehicle Detection and Segmentation from Aerial Images

Abstract

Image segmentation is crucial for computer vision. Visual segmentation simplifies image analysis. Detecting and dividing highways is a major difficulty in aerial traffic monitoring, autonomous driving, and border surveillance. This is tough. Image segmentation traditional algorithms are unsuccessful, according to the literature. Segmentation of the semantic field divides an image into semantically relevant components and assigns each to a class. Deep convolutional neural networks accurately segregate semantics. In deep learning, a convolutional neural network utilizes an input image to rate the relevance of various things. This work used convolutional neural networks to recognize and segment roads in aerial photos. SegNet and DeepLabv3+ use Vgg16 and ResNet-18 pre-trained models for road recognition and segmentation from aerial photos. SegNet based on Vgg16 produced high accuracy, whereas DeepLabv3+ based on Resnet-18 was efficient in terms of accuracy and training time. The suggested system designs use MATLAB. Learning and operational phases are included in the algorithm for image segmentation. In MATLAB, convolutional neural network analyses tagged aerial photographs. This work trains a convolutional neural network to accurately extract road features from aerial images. This system identifies roads and automobiles quickly for traffic monitoring.