Multi View 3D-Reconstruction

This repo contains the code for 3D reconstruction of objects from multiple images of different views. The implementation follows this post. Implementation is divided into following parts:

1. Feature Matching : In this section, each image is searched for interest points using Multilevel Harris Corner Detection and for each interest point a SIFT Descriptor is generated. For all possible pairs of images, these descriptors are crosschecked with each other to find the match between the correspoinding interest points. (This section needs more management)

2. Fundamental Matrix & Camera Matrix : It is assumed all the images are taken from same camera(that means intrinsic parameters will be same for all images). For all pair of images, a Fundamental Matrix F will be calculated using point correspondence generated in last section. Then, using intrinsic matrix K, we find Essential Matrix E = np.matmul(K.T, np.matmul(F, K)). Factorizing this E matrix we will get extrinsic matrix P = [R | t] of each image view.

3D.Reconstruction.Project.Output.mp4