A summary of 6D pose estimation
In the history of research on 6D pose estimation, a lot of different methods were invented with their own advantages and drawbacks. We now review existing work on 6D pose estimation from classical feature and template matching methods to newer end-to-end trainable CNN-based methods.
RGB:
Traditional RGB object instance recognition and pose estimation works used local keypoints and feature matching. Local descriptors needed by such methods were designed for invariance to changes in scale, rotation, illumination and viewpoints.Such methods are often fast and robust to occlusion and scene clutter. However, they only reliably handle textured objects in high resolution images.
RGB-D:
RGB-CNN:
RGB-D-CNN:
Others:
To be continued….
PREVIOUSWelcome