A project of the Robotics 2019 class of the School of Information Science and Technology (SIST) of ShanghaiTech University. Course Instructor: Prof. Sören Schwertfeger

Guoliang Gao and Zhouyang Lin


  There are xx stages we implemented in this project.

  At the first stage, we tried to use the Harris corner to mark the feature points in the single picture in order to achieve the function of optimal flow. But we found that the Harris corner function will mark the points outside the target we need to rebuild. So we got rid of the irrelevant point according to the sparsity of the feature points.

  As for the second stage, we use the optimal flow function on all image frames. When we was processing the image frames, we found that some tracked feature points would disappear along with the rotation of perspective. This will cause the failure of the reconstruction. So we had to remove the disappeared feature points from the matrix W, and track the new feature point.

  At the third stage, when working with the SVD decomposition, the calculation error cause the matrix Q not symmetry. Thus we might get the wrong result of matrix R and matrix S. To solve this problem, we estimate the probable error and erase them to get a symmetry matrix. Thus, we could reconstruct the model from the dataset.


  We write a c++ program to accomplish the feature points tracking, matrix W construction and its SVD decomposition to achieve the factorization.


Demo1: Using optimal flow to track the feature points from the dataset:

Demo2: Final resul: