Deep Visual Odometry With Adaptive Memory

We propose a novel deep visual odometry (VO) method that considers global information by selecting memory and refining poses. Existing learning-based methods take the VO task as a pure tracking problem via recovering camera poses from image snippets, leading to severe error accumulation. Global info...

Ausführliche Beschreibung

Bibliographische Detailangaben
Veröffentlicht in:IEEE transactions on pattern analysis and machine intelligence. - 1979. - 44(2022), 2 vom: 02. Feb., Seite 940-954
1. Verfasser: Xue, Fei (VerfasserIn)
Weitere Verfasser: Wang, Xin, Wang, Junqiu, Zha, Hongbin
Format: Online-Aufsatz
Sprache:English
Veröffentlicht: 2022
Zugriff auf das übergeordnete Werk:IEEE transactions on pattern analysis and machine intelligence
Schlagworte:Journal Article
LEADER 01000caa a22002652c 4500
001 NLM313252076
003 DE-627
005 20250227173638.0
007 cr uuu---uuuuu
008 231225s2022 xx |||||o 00| ||eng c
024 7 |a 10.1109/TPAMI.2020.3014100  |2 doi 
028 5 2 |a pubmed25n1044.xml 
035 |a (DE-627)NLM313252076 
035 |a (NLM)32749962 
040 |a DE-627  |b ger  |c DE-627  |e rakwb 
041 |a eng 
100 1 |a Xue, Fei  |e verfasserin  |4 aut 
245 1 0 |a Deep Visual Odometry With Adaptive Memory 
264 1 |c 2022 
336 |a Text  |b txt  |2 rdacontent 
337 |a ƒaComputermedien  |b c  |2 rdamedia 
338 |a ƒa Online-Ressource  |b cr  |2 rdacarrier 
500 |a Date Revised 10.01.2022 
500 |a published: Print-Electronic 
500 |a Citation Status PubMed-not-MEDLINE 
520 |a We propose a novel deep visual odometry (VO) method that considers global information by selecting memory and refining poses. Existing learning-based methods take the VO task as a pure tracking problem via recovering camera poses from image snippets, leading to severe error accumulation. Global information is crucial for alleviating accumulated errors. However, it is challenging to effectively preserve such information for end-to-end systems. To deal with this challenge, we design an adaptive memory module, which progressively and adaptively saves the information from local to global in a neural analogue of memory, enabling our system to process long-term dependency. Benefiting from global information in the memory, previous results are further refined by an additional refining module. With the guidance of previous outputs, we adopt a spatial-temporal attention to select features for each view based on the co-visibility in feature domain. Specifically, our architecture consisting of Tracking, Remembering and Refining modules works beyond tracking. Experiments on the KITTI and TUM-RGBD datasets demonstrate that our approach outperforms state-of-the-art methods by large margins and produces competitive results against classic approaches in regular scenes. Moreover, our model achieves outstanding performance in challenging scenarios such as texture-less regions and abrupt motions, where classic algorithms tend to fail 
650 4 |a Journal Article 
700 1 |a Wang, Xin  |e verfasserin  |4 aut 
700 1 |a Wang, Junqiu  |e verfasserin  |4 aut 
700 1 |a Zha, Hongbin  |e verfasserin  |4 aut 
773 0 8 |i Enthalten in  |t IEEE transactions on pattern analysis and machine intelligence  |d 1979  |g 44(2022), 2 vom: 02. Feb., Seite 940-954  |w (DE-627)NLM098212257  |x 1939-3539  |7 nnas 
773 1 8 |g volume:44  |g year:2022  |g number:2  |g day:02  |g month:02  |g pages:940-954 
856 4 0 |u http://dx.doi.org/10.1109/TPAMI.2020.3014100  |3 Volltext 
912 |a GBV_USEFLAG_A 
912 |a SYSFLAG_A 
912 |a GBV_NLM 
912 |a GBV_ILN_350 
951 |a AR 
952 |d 44  |j 2022  |e 2  |b 02  |c 02  |h 940-954