This project aims to implement SLAM using the input from a single Monocular camera.

There are plenty of different algorithms for MonoSLAM. The basic framework for this particular implementation is inspired by MonoSLAM: Real-Time Single Camera SLAM. However, this is not an exact replica of the framework proposed in that research work.

Only the particle filtering has been implemented as of now (reading pose estimates from groundtruth.txt)

1. Python 2.7.9
2. Some IDE for Python, preferably Spyder IDE. In Windows environment, Spyder is already included in WinPython For Mac and Linux, please consult the documentation
3. OpenCV needs to be installed and linked with Python in your system. Here are some useful links for that:

• OpenCV-Python interface in Windows
• Setting up OpenCV with Python support in Mac OSX
• OpenCV installation for Python in Ubuntu
• Python bindings for OpenCV in Debian

Download the code and open it up in Spyder (or anything else). Run the "Driver.py" script.

• To report any issues or bugs in the code, please use the Bug Tracker Form form.

• To view existing bugs, please check the Bug Tracker

This application is based on an extensive, offline dataset, created and maintained by Technische Universität München (TUM). The dataset contains hours of video captured by a depth-mapping camera. The dataset provides us with image sequence, as well as depth maps and unit quaternions describing the position and orientation of the camera w.r.t the world coordinates.

Link to RGB-D SLAM dataset.

[Detailed description of the dataset is available here.] (http://ais.informatik.uni-freiburg.de/publications/papers/sturm12iros.pdf)

Here, we use the image sequence to perform offline MonoSLAM and use the depth map + unit quaternion as ground truth for validating our algorithm.

We use the Sequence 'freiburg3_long_office_household' dataset, which has been captured using an Asus Xtion motion sensor.

After downloading the Sequence 'freiburg3_long_office_household' dataset from TUM's website, the following changes must be done to the source code:

• Open the Parameters.py file and set the value of the PATH variable to the folder containing the RGB images from the TUM dataset

• In the same Parameters.py file, set the GROUND_TRUTH_FILE variable to the text file containing the ground truth trajectory (this is required, as of now, since the SFM module have not been implemented yet)