Week 5 KinectFusion
KinectFusion is a very powerful tool that can help users create detailed 3D reconstructions indoor rapidly Even though the concept of depth camera is not new in this area, Kinect makes depth sensing popular due to its low-lost and real-time features.
This system allows people to use a handheld Kinect camera to move within a room and reconstruct a precise 3D model of the scene. The 3D model can be refined as users explore the space and new detailed data is added into the model. Besides, KinectFusion is not limited to scanning. It also does a lot of stuffs about augmented reality. Virtual objects can be registered spatially in the model as if it existed in real world. It is interesting that the virtual objects can also interact with the reconstructed scene to increase its sense of reality.
Despite the great success in this area, it is still a big challenge to create model of dynamic scenes. The authors tried to use an instance of ICP to predict the pose of foreground object and integrate the data into a second volume. In this way they succeeded in representing the entire foreground reconstruction.
Their future plan was to reconstruct larger scenes and explore fine-grained approaches of tracking and reconstruction.
Question to authors: Is there any trouble if people can not figure out what is real and what is virtual in such a model?
Comments are closed.