Visual data association in narrow-bandwidth networks

Tardioli, Danilo; Montijano, Eduardo; Mosteo, Alejandro R.

doi:10.1109/iros.2015.7353727

Cited by 18 publications

(18 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We prefer, instead, to use indirect relative measurements. These are established through registration of observations that two robots A and B make of the same scene [3], [23]. This requires no additional hardware: the sensors already used for odometry can be re-used for inter-robot measurements.…”

Section: B Decentralized Place Recognition and Pose Estimationmentioning

confidence: 99%

“…Another way to reduce bandwidth is to use compact representations that allow to establish inter-robot measurements with a minimal amount of data exchange. Visual feature-based maps can be compressed by compressing feature descriptors [27], substituting feature descriptors for corresponding cluster centers in a visual vocabulary [3], removing landmarks that prove not to be necessary for localization [28], [29] or using compact fullimage descriptors [30]. In previous work, we have shown another way to reduce the bandwidth of place recognition: data exchange can be reduced by a factor of the robot count if the problem can be cast to key-value lookup [31].…”

Section: B Decentralized Place Recognition and Pose Estimationmentioning

confidence: 99%

“…In this work, we use [1], an advanced version of [31] which at the same time makes use of a compact, state-of-the-art place representation, NetVLAD [30]. [3] is used in association of visual features for relative pose estimation.…”

Section: B Decentralized Place Recognition and Pose Estimationmentioning

confidence: 99%

“…However, it does not imitate [36] exactly, since the latter makes use of the data surrounding the match, which in decentralized RelPose would translate into a lot of data exchange. We want to avoid this and thus, we combine the relative pose estimation method form [36] with the visual data association method from [3], which reduces the data to be sent from α to β, d α→β RelPose , to a set of landmarks and visual word identifiers:…”

Section: Relative Pose Estimationmentioning

confidence: 99%

“…Both systems focus on data efficiency: they achieve performance similar to a centralized system while minimizing the data that needs to be exchanged. In this work, we integrate both systems along with a data-efficient method for visual feature association [3] into a full decentralized visual SLAM system. The system is evaluated on publically available data and the code is provided open-source.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Data-Efficient Decentralized Visual SLAM

Cieslewski

Choudhary

Scaramuzza

2018

2018 IEEE International Conference on Robotics and Automation (ICRA)

179

150

View full text Add to dashboard Cite

Decentralized visual simultaneous localization and mapping (SLAM) is a powerful tool for multi-robot applications in environments where absolute positioning systems are not available. Being visual, it relies on cameras, cheap, lightweight and versatile sensors, and being decentralized, it does not rely on communication to a central ground station. In this work, we integrate state-of-the-art decentralized SLAM components into a new, complete decentralized visual SLAM system. To allow for data association and co-optimization, existing decentralized visual SLAM systems regularly exchange the full map data between all robots, incurring large data transfers at a complexity that scales quadratically with the robot count. In contrast, our method performs efficient data association in two stages: in the first stage a compact full-image descriptor is deterministically sent to only one robot. In the second stage, which is only executed if the first stage succeeded, the data required for relative pose estimation is sent, again to only one robot. Thus, data association scales linearly with the robot count and uses highly compact place representations. For optimization, a state-of-the-art decentralized pose-graph optimization method is used. It exchanges a minimum amount of data which is linear with trajectory overlap. We characterize the resulting system and identify bottlenecks in its components. The system is evaluated on publically available data and we provide open access to the code.

show abstract

Section: B Decentralized Place Recognition and Pose Estimationmentioning

confidence: 99%