Learning to Segment Human Body Parts with Synthetically Trained Deep Convolutional Networks

Saviolo, Alessandro; Bonotto, Matteo; Evangelista, Daniele; Imperoli, Marco; Menegatti, Emanuele; Pretto, Alberto

doi:10.1007/978-3-030-95892-3_52

Cited by 4 publications

(2 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, these models require keypoint annotations to learn accurate pose encoding. Similarly, state-of-the-art human body segmentation neural networks [ 1 , 13 , 25 , 26 , 27 , 28 , 29 , 30 , 31 , 32 , 33 , 34 , 35 ] rely on bounding-box or pixel-level annotations for supervised body part segmentation. Given the absence of such annotations in the TSA datasets, we cannot directly use these supervised models.…”

Section: Related Workmentioning

confidence: 99%

Weakly Supervised 2D Pose Adaptation and Body Part Segmentation for Concealed Object Detection

Amadi

Agam

2023

Sensors

View full text Add to dashboard Cite

Weakly supervised pose estimation can be used to assist unsupervised body part segmentation and concealed item detection. The accuracy of pose estimation is essential for precise body part segmentation and accurate concealed item detection. In this paper, we show how poses obtained from an RGB pretrained 2D pose detector can be modified for the backscatter image domain. The 2D poses are refined using RANSAC bundle adjustment to minimize the projection loss in 3D. Furthermore, we show how 2D poses can be optimized using a newly proposed 3D-to-2D pose correction network weakly supervised with pose prior regularizers and multi-view pose and posture consistency losses. The optimized 2D poses are used to segment human body parts. We then train a body-part-aware anomaly detection network to detect foreign (concealed threat) objects on segmented body parts. Our work is applied to the TSA passenger screening dataset containing millimeter wave scan images of airport travelers annotated with only binary labels that indicate whether a foreign object is concealed on a body part. Our proposed approach significantly improves the detection accuracy of TSA 2D backscatter images in existing works with a state-of-the-art performance of 97% F1-score, 0.0559 log-loss on the TSA-PSD test-set, and a 74% reduction in 2D pose error.

show abstract

Section: Related Workmentioning

confidence: 99%

Weakly Supervised 2D Pose Adaptation and Body Part Segmentation for Concealed Object Detection

Amadi

Agam

2023

Sensors

View full text Add to dashboard Cite

show abstract

“…Our approach relies on a network of RGB-D cameras to be robust to occlusions [12]. The images acquired by each camera are segmented to recognise people and their body parts, and then projected from 2D to 3D exploiting the depth information acquired.…”

Section: Introductionmentioning

confidence: 99%

Multi-view Human Parsing for Human-Robot Collaboration

Terreran

Barcellona

Evangelista

et al. 2021

2021 20th International Conference on Advanced Robotics (ICAR)

Self Cite

View full text Add to dashboard Cite

In human-robot collaboration, perception plays a major role in enabling the robot to understand the surrounding environment and the position of humans inside the working area, which represents a key element for an effective and safe collaboration. Human pose estimators based on skeletal models are among the most popular approaches to monitor the position of humans around the robot, but they do not take into account information such as the body volume, needed by the robot for effective collision avoidance. In this paper, we propose a novel 3D human representation derived from body parts segmentation which combines high-level semantic information (i.e., human body parts) and volume information. To compute such body parts segmentation, also known as human parsing in the literature, we propose a multi-view system based on a camera network. People body parts are segmented in the frames acquired by each camera, projected into 3D world coordinates, and then aggregated to build a 3D representation of the human that is robust to occlusions. A further step of 3D data filtering has been implemented to improve robustness to outliers and segmentation accuracy. The proposed multi-view human parsing approach was tested in a real environment and its performance measured in terms of global and class accuracy on a dedicated dataset, acquired to thoroughly test the system under various conditions. The experimental results demonstrated the performance improvements that can be achieved thanks to the proposed multi-view approach.

show abstract

2D-Pose Based Human Body Segmentation for Weakly-Supervised Concealed Object Detection in Backscatter Millimeter-Wave Images

Amadi

Agam

2023

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Learning to Segment Human Body Parts with Synthetically Trained Deep Convolutional Networks

Cited by 4 publications

References 24 publications

Weakly Supervised 2D Pose Adaptation and Body Part Segmentation for Concealed Object Detection

Weakly Supervised 2D Pose Adaptation and Body Part Segmentation for Concealed Object Detection

Multi-view Human Parsing for Human-Robot Collaboration

2D-Pose Based Human Body Segmentation for Weakly-Supervised Concealed Object Detection in Backscatter Millimeter-Wave Images

Contact Info

Product

Resources

About