SLAM and NERF

Question 1

SLAM stands for

Answer 1

Simultaneous localization and mapping

Question 2

SLAM Methods

Answer 2

• Direct vs. Indirect
• Sparse vs. Dense
• Mono, Strereo, RGB-D
• Others: Laser, Sensor Fusion etc.

Question 3

3D Motion Estimation
* Given: 2 camera images
n point correspondences
* Wanted: Camera motion R,t (up to scale)

What are solutions

Answer 3

– 8-point algorithm
– normalized 8-point algorithm
– 6-point algorithm
– 5-point algorithm

Question 4

5 Point Algorithm assumption?

Answer 4

• Assumption: Calibrated Camera (K known), Otherwise: 8-Point Algorithm for Fundamental Matrix

Question 5

Estimate camera motion from frame to frame
* Problem:
– Estimates are inherently noisy
– Error accumulates over time -> drift

Solution?

Answer 5

Use loop-closures to minimize the drift / minimize the error over all constraints

Question 6

applications of SLAM in medical field

Answer 6

(Visual) Inside-Out Tracking
Localisation and Mapping in the OR

Question 7

what is Nerf

Answer 7

(NeRF) is a fully-connected neural network that can generate novel views of complex 3D scenes, based on a partial set of 2D images.

Question 8

NeRF steps

Answer 8

1. Marching camera rays through the scene to generate
   sampled set of 3D Points
2. Using these points and their 2D viewing directions as
   input to the neural network to get output as set of
   colors and densities
3. Using volume rendering to accumulate these colors and
   densities into a 2-D image

Question 9

NeRF Recipe

Answer 9

1. Intrinsic Calibration
2. Record Scene
3. Camera Pose Estimation
   a. Feature Extraction
   b. Feature Matching
   c. Triangulation
   d. Bundle Adjustment
4. NeRF Optimization
5. Novel View Rendering