Objective:

Vision Processing is known to take very high loads of processing power and with the increase in capability of embedded micro-controllers I am looking to investigate how to implement some simple vision based odometry on an embedded platform such as the X-MOS, mbed or alternatively create my own. In order to do this I will need to retain features (Lucas-Kanade) of two frames at one time, using Optical Flow, remove errors and anomalies or “outliers” and finally estimate the camera motion from the optical flow during this I will be required to maintain a good amount of features to track.

Reasons:

Foreseeable Pitfalls:

Project Plan:

• May 20th – Start
• May 31st – Get the C328 Camera working on the mbed (create easy to use functions which send a command to the camera and return a 1 if the camera responds with the correct ACK string, otherwise returns a 0)
• June 15th – Able to read and write to individual pixels in the RAW or JPEG file returned from the camera (create read/write functions which you pass the filename, x, y and variable to store the pixel data or value to change it to (respectively for read/write)) .
• June 30th -

Parts List:

• C328 JPEG Color Camera – Link
• mbed NXP LPC1768 – Link

Links and Similar Projects:

• “Two years of Visual Odometry on the Mars Exploration Rovers” – Link
• Human Tetris (Video Object Tracking) on an ATmega644 8-bit microcontroller – Link