ABSTRACT

This chapter analyzes several non-scene environmental factors and their effects on the loop closure detection algorithm based on image matching. It describes visual loop closure detection, which formulates a solution to the problem exploiting visual data, that is, using images captured by the robot, and it is the key step of the whole accumulated-error-elimination task. The chapter discusses a brief introduction to the related work on visual features for loop closure detection and provides the interference of non-scene factors to visual odometry calculation and image feature extraction. It presents the dataset with various non-scene factors and shows experimental results with dataset to compare the performance of two popular and successful loop closure detection algorithms. These are FAB-MAP and DBOW. The chapter explores the light, weather, pedestrian, vehicle factors of the real world environment as non-scene factors since they are nothing to do with scene-centric image matching which is the key step of loop closure detection.