This was project was difficult! It involved 360º images, equirectangular panoramas and plenty of epipolar geometric concepts to stitch images and find common points between different rooms of the houses. CreoTour is an startup that offer virtual tours services and they contact me to dramatically reduce the time of building a complete virtual tour.
As it always happens from theory to application things tend to get difficult while you advanced with different real examples. Basically because flats have very similar rooms and spaces or sometimes you can find funny things like mirrors.
After solving the initial idea using a deeplearning model (to solve Feature Matching and Homography) the project needed rules and reasoning to finally build automated virtual tours from the 360º images. To successfully finish a deeplearning project it’s always recommendable to use pandas and a little beat of data science, for this project it was the key to success.
Technology stack Python + deep learning + homography geometry + Pandas
