DR-SPAAM: A Spatial-Attention and Auto-regressive Model for Person Detection in 2D Range Data

Jia, Dan; Hermans, Alexander; Leibe, Bastian

doi:10.48550/arxiv.2004.14079

Cited by 1 publication

(9 citation statements)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Most recent developments [14], [3], [4] used deep learning techniques to detect persons directly from data, without manually engineered heuristics or features. The DROW detector [14] was the first deep learning-based walking aid detector working on 2D range data and was later extended to additionally detect persons [3].…”

Section: A Lidar-based Person Detectionmentioning

confidence: 99%

“…The DROW detector [14] was the first deep learning-based walking aid detector working on 2D range data and was later extended to additionally detect persons [3]. The current state-of-the-art method is the DR-SPAAM detector [4], which leveraged a temporal aggregation paradigm to incorporate multiple scans into the detection process, alleviating the problems associated with the low information content in a single LiDAR scan while retaining real-time computation on a mobile robotic platform.…”

Section: A Lidar-based Person Detectionmentioning

confidence: 99%

“…We experiment with two state-of-the-art person detectors, DROW3 [3] and its successor DR-SPAAM [4]. The DROW3 detector takes as input a 2D LiDAR scan, expressed as a onedimensional vector of range measurements.…”

Section: A Drow3 and Dr-spaam Detectormentioning

confidence: 99%

“…2D LiDARs provide accurate range measurements with a large field of view, often greater than 200 degrees, at an affordable price, and are a popular sensor choice for many robotic tasks, including person detection. While early approaches for detecting persons in 2D range data focused on heuristics with hand-crafted features [1], [2], recent studies used convolutional neural networks and further improved the detection results [3], [4].…”

Section: Introductionmentioning

confidence: 99%

“…Deep learning has become an integral part of modern detection algorithms, whether from 2D range data [3], [4] or images [5], [6], [7], [8], [9]. The success of these detectors hinges upon the availability of large and high-quality datasets.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Self-Supervised Person Detection in 2D Range Data using a Calibrated Camera

Jia¹,

Steinweg²,

Hermans³

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

Deep learning is the essential building block of state-of-the-art person detectors in 2D range data. However, only a few annotated datasets are available for training and testing these deep networks, potentially limiting their performance when deployed in new environments or with different LiDAR models. We propose a method, which uses bounding boxes from an image-based detector (e.g. Faster R-CNN) on a calibrated camera to automatically generate training labels (called pseudo-labels) for 2D LiDAR-based person detectors. Through experiments on the JackRabbot dataset with two detector models, DROW3 and DR-SPAAM, we show that selfsupervised detectors, trained or fine-tuned with pseudo-labels, outperform detectors trained using manual annotations from a different dataset. Combined with robust training techniques, the self-supervised detectors reach a performance close to the ones trained using manual annotations. Our method is an effective way to improve person detectors during deployment without any additional labeling effort, and we release our source code to support relevant robotic applications.

show abstract