Shang Yang

Vision navigation for aircrafts based on 3D reconstruction from real-time image sequences

In this paper, we propose a novel vision navigation method based on three-dimensional (3D) reconstruction from real-time image sequences. It adapts 3D reconstruction and terrain matching to establish the correspondence between image points and 3D space points and the terrain reference (by using a digital elevation map (DEM)). An adaptive weighted orthogonal iterative pose estimation method is employed to calculate the position and attitude angle of the aircraft. Synthesized and real experiments show that the proposed method is capable of providing accurate navigation parameters for a long-endurance flight without using a global positioning system or an inertial navigation system (INS). Moreover, it can be combined with an INS to achieve an improved navigation result.

Pose optimization based on integral of the distance between line segments

In this paper, new solutions for the problem of pose estimation from correspondences between 3D model lines and 2D image lines are proposed. Traditional line-based pose estimation methods rely on the assumption that the noises (perpendicular to the line) for the two endpoints are statistically independent. However, these two noises are in fact negatively correlated when the image line segment is fitted using the least-squares technique. Therefore, we design a new error function expressed by the average integral of the distance between line segments. Three least-squares techniques that optimize both the rotation and translation simultaneously are proposed in which the new error function is exploited. In addition, Lie group formalism is utilized to describe the pose parameters, and then, the optimization problem can be solved by means of a simple iterative least squares method. To enhance the robustness to outliers existing in the match data, an M-estimation method is developed to convert the pose optimization problem into an iterative reweighted least squares problem. The proposed methods are validated through experiments using both synthetic and real-world data. The experimental results show that the proposed methods yield a clearly higher precision than the traditional methods.

Research on CUDA-based image parallel dense matching

By analyzing the computer burden of image dense matching and the characteristic of Graphic Processor Unit(GPU) parallel computing mode, we designed a new solution of image parallel dense matching based on CUDA. We adopted the coarse-to-fine strategy, firstly implemented the pixel level normalized cross-correlation(NCC) matching method on CUDA; and then improved the matching precision by parallel affine least square matching(ALSM) under the GPU architecture. The proposed method implemented the dense matching in a full parallel mode, and took the advantage of the multi-threads supported by GPU, and finally obtained an obvious improvement in computational efficiency. The experiment results indicated that: the overall time consuming for the proposed dense matching method on GPU can achieve up to 25 times speedup over the version on CPU, which makes real-time 3D reconstruction become possible.