Convolutional Auto-Encoder and Independent Component Analysis Based Automatic Place Recognition for Moving Robot in Invariant Season Condition

• Published in: Human-Centric Intelligent Systems 2023, Vol.3 (1), p.13-24
• Main Authors: Islam, Md. Tariqul, Hasib, Khan Md, Rahman, Md. Mahbubur, Tusher, Abdur Nur, Alam, Mohammad Shafiul, Islam, Md. Rafiqul
• Format: Article
• Language: English
• Subjects: Auto-encoder; Computer Science; Deep learning; Independent Component Analysis; Machine learning; Principle Component Analysis; Research Article; SLAM
• ISSN: 2667-1336; 2667-1336
• DOI: 10.1007/s44230-022-00013-z

Building up a map is essential for mobile robots to localize their position and perfect autonomous navigation which is known as Simultaneous Localization and Mapping (SLAM). The map has become very important when the weather is inappropriate for the robot. However, the map becomes inconsistent when the robot moves in the environment and detects errors in its detection accuracy. The robot had difficulty identifying its previously visited path, which is called loop-closure detection when the climate changed immensely e.g. seasonal changes. The main goal of this work is to apply Independent Component Analysis (ICA) and Auto-Encoder (Convolutional Auto-Encoder and Fundamental Auto-Encoder) to understand the route through the robot. During the operation of robots across a wide range of environmental changing conditions, the ICA has auspicious potential to extract descriptors of condition-invariant images. On the other hand, Auto-Encoder has the capability to differentiate condition variant and condition invariant characteristics of a site and identify the most possible route for the robot. In order to complete this work perfectly, we used three seasonal datasets, they are Summer–Fall, Spring–Fall, and Summer–Spring datasets. This work uses the baseline method with a precision-recall curve and evaluates the performance of our proposed algorithm, especially the ICA algorithm. In short, the proposed algorithm ICA showed a 91.05% accuracy rate which is better than the baseline algorithm.