Publication

Abstract : Over a hundred thousand people are dying in road accidents in India every year, among those 30% of accidents take place due to roads and buildings which are constructed without following any government regulations. To minimize road accidents, we need to identify the buildings which are illegally constructed (without following any government regulations) on the roadside. The aim of this paper is divided into two parts: Firstly, segmenting the roads and buildings from the satellite image and secondly finding out the illegal construction of roads and buildings by matching the current image with the timed image which is authorized by the government authorities. To achieve the goal on a larger scale, a groundbreaking approach that combines a semantic segmentation neural network with U-Net has been developed. The convolution neural network is built using the U-Net architecture. In this application, we used the Massachusetts Benchmark dataset for roads and building extraction. Essentially, in this research, a comparison was carried out with other methods recently available (RESNET, SEGNET) and found out that this method has a greater advantage in providing results even with not fully connected layers like providing an efficient dense network which performs the prediction or identification efficiently. Moreover, this network was found to be equally good in terms of computational time.