The Future of Real-Time SLAM and Deep Learning vs SLAM
Part I: Why SLAM Matters
- SLAM is prime example of a what is called a "Geometric Method" in Computer Vision.
eg. CMU의 computer vision강의 = Learning-based Methods in Vision + Geometry-Based Methods in Vision 로 나누어짐
SLAM algorithms are complementary to ConvNets and Deep Learning:
- SLAM focuses on geometric problems
- Deep Learning is the master of perception (recognition) problems.
If you want a robot to go towards your refrigerator without hitting a wall, use SLAM. If you want the robot to identify the items inside your fridge, use ConvNets.
- SLAM is a real-time version of Structure from Motion (SfM)
- Structure from Motion vs Visual SLAM
- Structure from Motion (SfM) and SLAM are solving a very similar problem,
- SfM is traditionally performed in an offline fashion
- SLAM has been slowly moving towards the low-power / real-time / single RGB camera mode of operation.
- SfM 문제점 : 큰 구조물은 많은 사진이 필요 하고 처리 하는데 많은 시간이 걸림.
given a large collection of photos of a single outdoor structure (like the Colliseum), construct a 3D model of the structure and determine the camera's poses. The image collection is processed in an offline setting, and large reconstructions can take anywhere between hours and days.
- sfM 관련 소프트웨어 라이브러리
- Bundler, an open-source Structure from Motion toolkit
- Libceres, a non-linear least squares minimizer (useful for bundle adjustment problems)
- Andrew Zisserman's Multiple-View Geometry MATLAB Functions
Part II: The Future of Real-time SLAM
- MonoSLAM (2003년 Andrew Davison 주도)
- PTAM
- FAB-MAP
- DTAM
- KinectFusion
- Talk 1: Christian Kerl on Continuous Trajectories in SLAM
- Talk 2: Semi-Dense Direct SLAM by Jakob Engel
생략 ...
Part III: Deep Learning vs SLAM
workshop presenters agreed that semantics are necessary to build bigger and better SLAM systems
- Integrating semantic information into SLAM
- "Will end-to-end learning dominate SLAM?"