Part 1: Introduction & Feature Detection

Some people claim to have a poor sense of direction. They might compare themselves to a robot and feel much better.

Part 2: Pose Optimization

Once our features are detected and associated, we aim to estimate the movement between the two observations.

After Part 1, which focused on feature detection and association, we now delve into estimating our position and orientation over time.

Choosing the Pose Model to Optimize

A pose represents the position and orientation of our robot in space. We can express it in the way and in the coordinate system that suits us best (I’ve seen spherical, but here, we prefer Cartesian).

Part 2: Improvements and More Blabla!

We left off with the construction of a basic search tree structure using a possibility-based search with a balance of exploration, score, and chance (MCTS: Monte Carlo Tree Search). In this second part, I will introduce the improvements I found for the basic method.

Part 1: Blabla and Implementation

I’m terrible at chess, checkers, get rolled over in Connect 4 – in short, board games are not my strong suit. However, I do fairly well in programming.

In this post, I implement a real time depth map segmentation system, based on normal map analysis.

Use a video taken by a single camera to estimate the depth of objects in an image. A small dip in the world of epipolar geometry and key points analysis.

Certainly! Here is the translation:

CAPE (Cylinder And Plane Extraction) is an extremely efficient method for extracting planes and cylinders in RGB-D images, based on an AHC (Agglomerative Hierarchical Clustering) method. Despite the effectiveness of this method, the C++ implementation of the paper has many flaws.

In this post, I will explain how I fixed most of these issues.

Localize a mobile robot in a virtual environment with constraint programming.