Tatoglu, Akin. (Stevens Institute of Technology) “Modified monocular SLAM with concurrent model parameter identification”, (2015)

You are here

Inside Signal Processing Newsletter Home Page

Top Reasons to Join SPS Today!

1. IEEE Signal Processing Magazine
2. Signal Processing Digital Library*
3. Inside Signal Processing Newsletter
4. SPS Resource Center
5. Career advancement & recognition
6. Discounts on conferences and publications
7. Professional networking
8. Communities for students, young professionals, and women
9. Volunteer opportunities
10. Coming soon! PDH/CEU credits
Click here to learn more.

10 years of news and resources for members of the IEEE Signal Processing Society

Tatoglu, Akin. (Stevens Institute of Technology) “Modified monocular SLAM with concurrent model parameter identification”, (2015)

Tatoglu, Akin. (Stevens Institute of Technology) “Modified monocular SLAM with concurrent model parameter identification”, (2015) Advisor: Pochiraju, Kishore

In order to navigate autonomously, mobile robots require an awareness of their position and a map of the surroundings. Determination of a robot's position in a map is called localization. In many cases a map of the surroundings may not exist or may be unreliable. Simultaneous Localization and Mapping (SLAM) algorithms determine both the position estimate and a map of the surroundings as the robot traverses its environment. SLAM considers the control input that directs the robot motion and the information from sensors such as a laser ranger, stereo camera or wheel rotation encoders to stochastically minimize the errors in position and map estimates. Monocular SLAM or MonoSLAM attempts the localization and mapping tasks with a single camera as the sensor. Challenges for generating accurate position and map information with MonoSLAM are numerous. Calibration of the depth-scaling of the camera, choice of the right motion and noise models, and the quality of the image generated by the sensor are major issues that need to be addressed before MonoSLAM can be effectively implemented.

In this thesis the effectiveness of MonoSLAM as a localization algorithm during navigation of mobile robots has been evaluated with a multi-ball drive mobility platform, as well as with robotic arm and linear actuation robot. Mapping and trajectory accuracies are investigated under various motion profiles. MonoSLAM algorithms have been extended by including real-time identification of motion and noise models. As MonoSLAM contains no information about the motion of the robot, a constant velocity model with acceleration noise is generally assumed. The location prediction from MonoSLAM is found to drift considerably during motion profile changes. This drift is attributed to artificial movement of landmarks due to the errors in Kalman gain computation introduced by the motion assumptions. In this effort, both motion and noise models are identified by an optical flow pre-processor. Optical flow analyzes the image stream for changes in robot dynamics and identifies an appropriate motion model used in localization. Another methodology based on Pearson product moment correlation (Pearson-R) is employed to detect motion behavior changes. In both methods, detected motion profiles are used to construct appropriate motion and noise models for localizing robotic machinery and mobile robots.

A modified SLAM algorithm, termed the CMPISLAM (SLAM with Concurrent Model Parameter Identification), is hypothesized and its performance has been evaluated. CMPISLAM introduces a pipelined structure in which two simultaneous analytical methods operate on a single stream of the image data. The position and map information updates after a short time delay required for identifying an appropriate motion model. Test cases are formulated to benchmark the performance of MonoSLAM and CMPISLAM under various motion profiles. The main contribution of this thesis is that the concurrent identification of motion model enables the use of MonoSLAM and its variants for robot navigation mitigating the need for sensors.

Table of Contents:

SPS on Twitter

SPS Videos


Signal Processing in Home Assistants

 


Multimedia Forensics


Careers in Signal Processing             

 


Under the Radar