Scene Modeling through Multi-modal Geospatial Data Fusion
Abstract
Precise and realistic scene modeling is a key need for many Army endeavors and operations. Conventional ways to carry out this task are through geospatial sensing, either from overhead or ground. However, due to occlusion and access restrictions, either of the two sensing approaches alone cannot fulfill this critical need. Fusion is therefore needed to combine such multi-modal sensing data collected from different platforms at varying distances. Due to the growing popularity of low-cost imaging or ranging sensors, e.g., the ones embedded in handheld smartphones or aboard off-the-shelf UAVs, as well as the public availability of such ubiquitous sensing data, we propose to integrate them with overhead sensing data to achieve a complete and realistic scene modeling. Our effort intends to address the fundamental difficulties in such endeavor. The proposed work consists of four challenging tasks referring to multi-modal geospatial data fusion. First, we will optimize image networking and triangulation to precisely and reliably determine the sensor geometry and locations. Our second task is to create a point cloud representation of the scene by an improved dense matching for ground level images. The next task is to accurately combine overhead point clouds with the ground ones to form a complete, coherent scene representation. Finally, interest objects will be detected and modeled from the combined point clouds under the constraints of geometric correctness and topologic consistence. The intrinsic challenge of the proposed study is due to the co-existence of large redundant sensing on one hand and large incomplete coverage on the other. We expect our work would promote the general concept of citizen science and open science in the ever growing geospatial domain. Our study will not only provide a theoretical solution to the technical difficulties in the domain-specific problem but also demonstrate the capability and potential of citizen data in a broader scope. The bulk of our work will be based on open source tools and some of our outcome will also be shared as open source.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 11, 2018
- Source ID
- W911NF1710404
Entities
People
- Jie Shan
Organizations
- Army Contracting Command
- United States Army
- University of Virginia