China
Africa
Explore chapters and articles related to this topic
Recognition of Emotions in the Elderly through Facial Expressions: A Machine Learning-Based Approach
Published in Wellington Pinheiro dos Santos, Juliana Carneiro Gomes, Valter Augusto de Freitas Barbosa, Swarm Intelligence Trends and Applications, 2023
Arianne Sarmento Torcate, Maíra Araújo Santana, Juliana Carneiro Gomes, Ingrid Bruno Nunes, Flávio Secco Fonseca, Gisele M.M. Moreno, Wellington Pinheiro dos Santos
The study carried out by Kuruvayil and Palaniswamy (2021) seeks to investigate the challenges regarding the extraction of facial features. Some of these challenges are partial occlusion, pose and lighting variations. Therefore, the authors propose a modeling of an emotion recognition system based on machine learning and deep learning that aims to generalize well in natural obstructions. The training was carried out using a large volume of data. One challenge was the scarcity of images with the desired characteristics in the basic emotion recognition datasets. The proposed system, ERMOPI (Emotion Recognition using Meta-learning through Occlusion, Pose and Illumination), was trained for 5 basic emotions with facial images having 5 occlusion categories, 7 head poses and 5 lighting levels. The results were 90% accuracy for CMU Multi-PIE images (dataset) and 68% accuracy for AffectNet images (dataset).
Standard Ontologies and HRI
Published in Paolo Barattini, Vicentini Federico, Gurvinder Singh Virk, Tamás Haidegger, Human–Robot Interaction, 2019
Sandro Rama Fiorini, Abdelghani Chibani, Tamás Haidegger, Joel Luis Carbonera, Craig Schlenoff, Jacek Malec, Edson Prestes, Paulo Gonçalves, S. Veera Ragavan, Howard Li, Hirenkumar Nakawala, Stephen Balakirsky, Sofiane Bouznad, Noauel Ayari, Yacine Amirat
In POS, a position and an orientation constitute a pose. The pose of an object is the description of any position and orientation simultaneously applied to the same object. Often, a pose is defined with a position and an orientation referenced to different coordinate systems/reference objects. In addition, since objects can have many different positions and orientations, they can also have many different poses.
Basics of Image Processing
Published in Maheshkumar H. Kolekar, Intelligent Video Surveillance Systems, 2018
Pose estimation: This is a technique of estimating the position or orientation of a specific object relative to the camera. For example, the application of assisting a robot arm in retrieving objects from a conveyor belt.
HAR-time: human action recognition with time factor analysis on worker operating time
Published in International Journal of Computer Integrated Manufacturing, 2023
Chao-Lung Yang, Shang-Che Hsu, Yu-Wei Hsu, Yu-Chung Kang
Essentially, for an image or a video, the human skeleton information can be retrieved using human pose estimation technology. Human pose estimation is to detect the parts such as the joints, arms, or neck of human body given a human image. Due to the emerging of deep-learning network, the accuracy of pose estimation has been improved dramatically. Starting from utilizing deep-learning methodology on the single-person pose estimation, Toshev et al. first used the AlexNet architecture named as DeepPose to directly regress spatial joint coordinates (Toshev and Szegedy 2014). Afterward, the deep neural networks-based models have quickly dominated the pose estimation problem (Kocabas, Karagoz, and Akbas 2018). Recently, an innovative method UniPose proposed by Artacho et al. which combines CNN architecture and LSTM, with ResNet backbone and Waterfall module, shows promising performance on single person pose detection for both single images and videos (Artacho and Savakis 2020). Xiao et al. proposed baseline methods for pose estimation and tracking (Xiao, Wu, and Wei 2018). By adding a few deconvolutional layers into the backbone network for pose estimation and utilizing optical flow-based pose propagation and similarity measurement for pose tracking, the human pose could be estimated more effectively.
Assessment of deep learning pose estimates for sports collision tracking
Published in Journal of Sports Sciences, 2022
Richard Blythman, Manan Saxena, Gregory J. Tierney, Chris Richter, Aljosa Smolic, Ciaran Simms
The LT model outputs a skeleton configuration with 17 joints (as shown by the combined set of red and blue dots in Figure 1(b)), since it is trained on the Human3.6 M dataset. However, the original dataset provides no information about the method used for fitting the skeleton to the markers. For example, it is not clear how joints for the spine and neck are estimated, which typically require extra markers (e.g., on the abdomen). Hence, we choose to exclude these from comparison with our staged tackle dataset. The resulting skeleton model of 15 joints is shown by the set of red dots in Figure 1(b). The MS COCO (Lin et al., 2015) and MPII (Andriluka et al., 2014) 2D datasets are also used to pre-train the backbone of the LT model for 2D human pose estimation. The 16-joint skeleton model of MPII is most similar to the Human3.6 M skeleton, using all of the same joints except for the spine joint (and having small differences in the position of the thorax and neck). We leave out the neck joint to give us a matching set of 15 joints for the 2D dataset. The skeleton model of the COCO dataset contains 17 joints with 5 joints located on the head and face but no pelvis and spine joint. However, the compatibility is less important here since the 2D model is re-trained on the MPII dataset.
Reconsideration of multi-stage deep network for human pose estimation
Published in Computer Methods in Biomechanics and Biomedical Engineering: Imaging & Visualization, 2021
Pratishtha Verma, Rajeev Srivastava
Computer vision is a versatile field related to artificial intelligence, machine learning, robotics, and signal processing. The purpose of computer vision is to program a computer to understand, process and analyse images. The three fundamental tasks of computer vision are object detection and pose estimation, tracking and segmentation. Object detection and pose estimation are some of the essential jobs of computer vision due to their wide range of applications. Till date, Face and Human are the most examined objects. Human beings are usually considered as an articulated subject consisting of fixed moving parts linked to certain expressive points. Under this assumption, HPE is the technique of deriving an exact pixel position from human body key points. It is an essential tool to solve other high-level tasks, such as human tracking, action recognition, human-computer interaction, motion capture, content retrieval, social signal processing, and animation (Belagiannis and Zisserman 2017).