Change detection and the updating of SLAM maps

• Main Author: Foong, Kok Leong
• Other Authors: Seet Gim Lee, Gerald
• Format: Final Year Project
• Language: English
• Published: 2018
• Subjects: DRNTU::Engineering
• Online Access: http://hdl.handle.net/10356/75141
• Institution: Nanyang Technological University

Overview – The food and beverage industry has become increasingly competitive. It is customary for restaurants to operate from morning till night to earn more profit. The seating arrangement in restaurant is always reconfigured. It will be convenient if a robot can automate these tasks, so that the time spent and injuries associated with the manual handling of seating arrangement can be reduced. The robot requires a map to locate the position of tables and chairs. This map is constructed by sending a robot into the venue to perform Simultaneous Localization and Mapping (SLAM) of the unknown environment. Problem – Whenever there is a change in the environment, the robot has to re-map the changed environment again. This process is repetitive and time-consuming. In addition, sensor noises, and the computational efficiency and pose estimation accuracy of SLAM algorithms are the issues as well. Solution – Hence, it was proposed to install an overhead camera to capture these changes. This approach is convenient as the robot does not have to perform remapping. Any changes are captured by the camera, processed and updated onto the previously constructed SLAM map. This project aims to develop a software program to carry out change detection of tables and chairs. The change detection refers to the change in centroid coordinates and angular orientation of tables and chairs. These changes are updated onto a SLAM map, which are useful by a robot to pinpoint objects locations and carry out objects removal tasks. Approaches – Two approaches were developed and investigated: the color approach and the features detection approach, in terms of their invariance and stability to illumination, scale, rotation and 3D point of view from different angles. Different image processing algorithms from OpenCV-3.2.0 and opencv_contrib libraries were used in both approaches. Basically, color approach makes use of color filter, whereas features detection approach makes use of template matching, Oriented FAST and Rotated BRIEF (ORB) feature detector, and Shi-Tomasi Corner Detector. Conclusion – This report concludes that features detection approach be used in table detection and localization because it achieved 87% success rate in detecting table corners and location. The unsuccessful instances are attributed to objects of similar color with table and change in 3D point of view from different angles, which caused template matching to fail. Due to the lack of features in chairs, color approach was used in chair detection and localization.