Research Projects

The United Nations' Sustainable Development Goals aim to improve the accessibility of emergency services, particularly in dense cities like India. A geospatial inventory of road attributes, including width and presence of objects like electric wires, can help in preventive planning and resource forecasting. However, current research focuses on extracting road features through rule-based algorithms, with less attention given to quantifying road width estimation through deep learning. This study aims to fill this gap by developing new CNN architectures that train on a manually labeled database. The proposed approach combines ultra-high-resolution photographic images with Light Ranging and Detection LiDAR collected from airborne and mobile platforms. The extracted road objects are vectorized to calculate width, and an algorithm is developed to deduce the location, shape, and size of the extracted road features. The research will be the first of its kind in integrating spatial data modeling, machine learning, 2D/3D GIS, high-resolution remote sensing, and urban analytics to develop a geospatial inventory of vehicle access in a city. This research can be extended from local to regional and national levels, reducing effort while improving existing methods. The project aims to develop a labelled imagery and LiDAR database of urban built-up, train CNN models to extract road features, develop an algorithm to deduce road width from segmented road features, create a geospatial inventory of roads based on emergency vehicle details, design accessibility maps for different vehicle classes, and prepare a standard operating procedure for creating such maps. The project is innovative in generating an ensembled labelled database of built-up features from images and LiDAR data captured from aerial and mobile platforms. The project also aims to develop an automated geospatial inventory of vehicle access in cities considering vehicle properties. The Co-PI has been involved in high-resolution data generation for the city of Chandigarh from airborne and mobile platforms, including LiDAR point cloud and photographic images. The methodology involves training a deep learning model using fused photographs and LiDAR datasets, manual labelling, and a Convolution Neural Network CNN model.