Title:  A Multi-view Stereo Approach for Radar Based Shape Inversion

Committee: 

Dr. Yezzi, Advisor        

Dr. Barnes, Chair

Dr. Lanterman

Abstract:

The objective of the proposed research is to develop a novel method for dense surface reconstruction of scenes using radar. For a given scene and a set of antennas looking towards to this scene, our method estimates the shape of the scene using the radar return signal. For this purpose, we use a deformable shape evolution approach which seeks to match the received signal to a computed forward model based on the evolving shape. Using such an approach comes with important advantages such as the ability to model the issues related to the object geometry which cannot be easily incorporated into the problem by the current radar based imaging techniques. As an example, we know that most scene surfaces have some level of smoothness. Exploiting such prior information can yield a more accurate estimation of the shape. It can also decrease the number of measurements required for an accurate estimation since the prior information limits the solution space to a subspace that favors surface smoothness. Another important geometric consideration is the self-occlusions present in the scene. We know that certain parts of the object surface are not visible for some antenna positions which can be very important to model, especially for close range applications in which the self-occlusions strongly change with the viewpoint. Iterations start with an initial shape which is gradually deformed until its image under the forward model gets sufficiently close to the actual measured signal. However, using an iterative inversion scheme for radar can be tricky as radar signals are highly oscillatory with respect to the surface shape which can introduce itself in the cost functional if cost function is not carefully designed. For this purpose, we employ the technique of stretch processing to extract geometric properties of the shape from radar return signal. This yields a smooth and geometric cost functional by which shape inversion can be robustly performed via gradient-based minimization algorithms. Synthetic simulations with a polygonal shape model show the promise of  our approach on some challenging shapes.