Real-time Dense Point Cloud, Digital Surface Map (DSM) and (Ortho-)Mosaic Generation for UAVs

Official implementation of our ICRA'22 paper: SPIN Road Mapper: Extracting Roads from Aerial Images via Spatial and Interaction Space Graph Reasoning for Autonomous Driving

UAV-Mapper is a lightweight UAV Image Processing System, Visual SFM reconstruction or Aerial Triangulation, Fast Ortho-Mosaic, Plannar Mosaic, Fast Digital Surface Map (DSM) and 3d reconstruction for UAVs.

DeepAerialMapper is an open-source project to semi-automatically convert segmentation masks of aerial imagery into High-Definition maps (HD maps).

Mapping aerial lidar data using ros

A software package is built for display and classification of Hyperspectral Images captured byIMS-1 HySI sensor has been developed using SAM.The construction and display of the 3-D cube by considering all the 64 bands of image at a time. The identification of classes in the Hyperspectral Image using a supervised classification algorithm called the Spectral Angle Mapper Algorithm. Results are obtained to read and reorganize multiple 2-D datasets into a single compact 3D dataset cube.Thematic Information Extraction — Supervised Classification Remotely sensed data may be analyzed to extract use- ful thematic information. This transforms the data into in- formation. For example, themes may include land-cover, water bodies, and clouds. The classification may be per- formed using supervised, unsupervised and fuzzy set clas- sification approaches. In a supervised image classification, the identity and lo- cation of some of the land-cover types should be known beforehand through a combination of fieldwork, analy- sis of aerial photography, maps, and personal experience. The analyst attempts to locate sites in the remotely sensed data that represent homogeneous examples of these known land-cover types. These areas are commonly referred to as training sites because the spectral characteristics of these known areas are used to train the classification algorithm for eventual land-cover mapping of the remainder of the image. Multivariate statistical parameters such as means, standard deviations, and covariance matrices are calculated for each training site. Every pixel both inside and outside these training sites is then evaluated and assigned to the class where it has the highest likelihood of being a mem- ber. This is often referred to as hard classification because a pixel is assigned to only one class (e.g., forest), even though the sensor records the radiant flux from a mixture of biophysical materials, for example: 10% bare soil, 20% scrub shrub, 70% forest.

No description available

Aerial Vehicle Mapper is an AI-powered tool that detects and maps vehicles from aerial images using the Detectron2 deep learning framework. It enables geospatial mapping of detected vehicles for applications in urban planning, surveillance, disaster management, and more.