3D reconstruction of freely moving persons for re-identification with a depth sensor

Munaro, Matteo; Basso, Alberto; Fossati, Andrea; Gool, Luc J. Van; Menegatti, Emanuele

doi:10.1109/icra.2014.6907518

Cited by 86 publications

(84 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Besides that, Gillian and Paradiso [7] intruduce the Gesture Recognition Toolkit that was designed to enable even non-specialist users to work on realtime machine learning and gesture recognition. Munaro et al [16] use the point cloud of a consumer depth sensor to create a 3D-model of a person for re-identification. Skeletal data is utilized to overcome the problem of comparing different poses.…”

Section: Multi-depth-camera Skeleton Tracking System Use Casesmentioning

confidence: 99%

FusionKit

Rietzler

Geiselhart

Thomas³

et al. 2016

Proceedings of the 8th ACM SIGCHI Symposium on Engineering Interactive Computing Systems

View full text Add to dashboard Cite

We present a toolkit for markerless skeleton tracking and marker-based object tracking utilizing data fusion with an arbitrary number of depth cameras. As depth-camera based skeletal tracking is always inaccurate due to technology limitations, our goal was to be able to preestimate systematic errors for given tracking situations to improve fusion. Previous work analyzed various aspects of depth camera accuracy, however to our best knowledge, there has been neither systematic error modelling nor an application of such a model for skeletal fusion. Our paper presents such a model for the Kinect v2 camera, by using statistical modelling on capture datasets using such cameras and a marker-based ground truth capture system. By applying this model, we are able to improve the overall accuracy of the fusion output by 68% by predicting data quality with an error of around 3.2 cm. Our toolkit is available for use by other researchers to easily create larger capture spaces with higher tracking accuracy based on the error model when compared to single depth cameras.

show abstract

Section: Multi-depth-camera Skeleton Tracking System Use Casesmentioning

confidence: 99%

FusionKit

Rietzler

Geiselhart

Thomas³

et al. 2016

Proceedings of the 8th ACM SIGCHI Symposium on Engineering Interactive Computing Systems

View full text Add to dashboard Cite

show abstract

“…After the release of the Microsoft Kinect in 2010, several RGB-D datasets were published. RGB-D datasets for human recognition were also provided, such as for the reidentification of a person with RGB-D sensors [8], BIWI RGBD-ID dataset [9], and UPCV Gait dataset [10]. As the Kinect cannot measure depths greater than 10 [m], LiDAR sensors were employed to handle depths over 10 [m].…”

Section: Related Workmentioning

confidence: 99%

Automatic Labeled LiDAR Data Generation based on Precise Human Model

Kim

Tanaka

Okutomi

et al. 2019

2019 International Conference on Robotics and Automation (ICRA)

View full text Add to dashboard Cite

Following improvements in deep neural networks, stateof-the-art netwy depends on the training data. An issue with collecting training data is labeling. Labeling by humans is necessary to obtain the ground truth label; however, labeling requires huge costs. Therefore, we propose an automatic labeled data generation pipeline, for which we can change any parameters or data generation environments. Our approach uses a human model named Dhaiba and a background of Miraikan and consequently generated realistic artificial data. We present 500k+ data generated by the proposed pipeline. This paper also describes the specification oforks have been proposed for human recog-nition using point clouds captured by LiDAR. However, the performance of these networks strongl the pipeline and data details with evaluations of various approaches.

show abstract

“…Depth data has been widely used in gesture and action recognition [2][3][4]17]. While computer vision has enabled many novel and exciting insights into sports performance, there are other instances where vision alone is insufficient for extract meaningful performance features, and in those instance 3D data may provide a practical solution.…”

Section: Introductionmentioning

confidence: 99%

Non-Invasive Performance Measurement in Combat Sports

Behendi

Morgan

Fookes

2015

Proceedings of the 10th International Symposium on Computer Science in Sports (ISCSS)

View full text Add to dashboard Cite

Abstract.Computer vision offers a growing capacity to detect and classify actions in a large range of sports. Since combat sports are highly dynamic and physically demanding, it is difficult to measure features of performance from competition in a safe and practical way. Also, coaches frequently wish to measure the performance characteristics of other competitors. For these reasons it is desirable to be able to measure features of competitive performance without using sensors or physical devices. We present a non-invasive method for extracting pose and features of behaviour in boxing using vision cameras and time of flight sensors. We demonstrate that body parts can be reliably located, which allow punching actions to be detected. Those data can then visualised in a way that allows coaches to analysis behaviour.

show abstract

3D reconstruction of freely moving persons for re-identification with a depth sensor

Cited by 86 publications

References 17 publications

FusionKit

FusionKit

Automatic Labeled LiDAR Data Generation based on Precise Human Model

Non-Invasive Performance Measurement in Combat Sports

Contact Info

Product

Resources

About