Object_ORB_SLAM2

Fork from ORB_SLAM2, added the object detection, the object based 2D-3D point macthing and object based map point creation, so that the object can be generate from 2D image to 3D points cloud.

1. Object Detection

Use Mask-RCNN to detect objects for each keyframes.

2. 2D Keypoint Generation

Give each keypoint a new attribute named object Id, which is connected to the object it belongs to.

3. 3D Map Point Cloud Generation

For each object detected by Mask-RCNN in each keyframe, generateing the 3D map points based on 2D keypoint with object Id. So that the 3D points in the map are no longer the meanless sparse point, but a part of the object point cloud. After we get the point cloud for objects, we can do more interactions with objects in AI(like AR, Auto Driving). We can see in the image above, red points represent the monitor, green points stand for the keyboard, blue points stand for the plant and yellow points stand for the desk.