Methods for Appearance-based Loop Closure Detection by Emilio Garcia-Fidalgo & Alberto Ortiz

Author:Emilio Garcia-Fidalgo & Alberto Ortiz
Language: eng
Format: epub
Publisher: Springer International Publishing, Cham

(4.1)

(4.2)

Informally, a high precision value means a low number of false positives, while a high recall value means a low number of false negatives. A graphical representation of the precision and recall metrics is shown in Fig. 4.1.

In order to use the precision-recall metrics to evaluate the performance of the algorithms presented in this book, the datasets introduced in Sect. 4.2 are provided with a ground truth, which indicates, for each image in the sequence, which other images can be considered to close a loop with it. Then, the assessment is performed counting for each dataset the number of true positives, true negatives, false positives and false negatives, where positive is meant for detection of loop closure. The precision P is then defined as the ratio of real loop closure detections to total amount of loop closures detected, while the recall R is defined as the ratio of real loop closures to total amount of loop closures existent in the dataset.

For loop closure detection, it is essential to avoid false positives, since it means that two images have been identified as a loop but, in reality, they represent different places. This will induce the algorithm to produce inconsistent maps and, therefore, avoiding these false positives becomes essential. By definition, if no false positives are found, the precision reaches 100% (see Eq. 4.1). Then, in our experiments, we are interested in finding the best recall than can be achieved at 100% of precision using each approach, which indicates the percentage of loop closures that can be detected by the algorithm without false positive detections.

Fig. 4.1Precision and recall metrics. The items that are positive according to the ground truth are located to the left of the straight line, while the items retrieved as positive by the binary classifier are inside the oval area. The red areas represent errors. Then, the red area located to the left of the line and outside of the oval area represents the positive items that could not be retrieved (false negatives), while the red area inside the oval area represents the items retrieved as positives that are not actually positives (false positives)

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.