PGNet: Pipeline Guidance for Human Key-Point Detection

Keypoint Detection for Injury Identification during Turkey Husbandry Using Neural Networks

Injurious pecking against conspecifics is a serious problem in turkey husbandry. Bloody injuries act as a trigger mechanism to induce further pecking, and timely detection and intervention can prevent massive animal welfare impairments and costly losses. Thus, the overarching aim is to develop a camera-based system to monitor the flock and detect injuries using neural networks. In a preliminary study, images of turkeys were annotated by labelling potential injuries. These were used to train a network for injury detection. Here, we applied a keypoint detection model to provide more information on animal position and indicate injury location. Therefore, seven turkey keypoints were defined, and 244 images (showing 7660 birds) were manually annotated. Two state-of-the-art approaches for pose estimation were adjusted, and their results were compared. Subsequently, a better keypoint detection model (HRNet-W48) was combined with the segmentation model for injury detection. For example, individual injuries were classified using “near tail” or “near head” labels. Summarizing, the keypoint detection showed good results and could clearly differentiate between individual animals even in crowded situations.

Moving towards intelligent telemedicine: Computer vision measurement of human movement

Background:

Telemedicine video consultations are rapidly increasing globally, accelerated by the COVID-19 pandemic. This presents opportunities to use computer vision technologies to augment clinician visual judgement because video cameras are so ubiquitous in personal devices and new techniques, such as DeepLabCut (DLC) can precisely measure human movement from smartphone videos. However, the accuracy of DLC to track human movements in videos obtained from laptop cameras, which have a much lower FPS, has never been investigated; this is a critical gap because patients use laptops for most telemedicine consultations.

Objectives:

To determine the validity and reliability of DLC applied to laptop videos to measure finger tapping, a validated test of human movement.

Method:

Sixteen adults completed finger-tapping tests at 0.5 Hz, 1 Hz, 2 Hz, 3 Hz and at maximal speed. Hand movements were recorded simultaneously by a laptop camera at 30 frames per second (FPS) and by Optotrak, a 3D motion analysis system at 250 FPS. Eight DLC neural network architectures (ResNet50, ResNet101, ResNet152, MobileNetV1, MobileNetV2, EfficientNetB0, EfficientNetB3, EfficientNetB6) were applied to the laptop video and extracted movement features were compared to the ground truth Optotrak motion tracking.

Results:

Over 96% (529/552) of DLC measures were within +/−0.5 Hz of the Optotrak measures. At tapping frequencies >4 Hz, there was progressive decline in accuracy, attributed to motion blur associated with the laptop camera’s low FPS. Computer vision methods hold potential for moving us towards intelligent telemedicine by providing human movement analysis during consultations. However, further developments are required to accurately measure the fastest movements.

Hand Gesture Recognition Based on Auto-Landmark Localization and Reweighted Genetic Algorithm for Healthcare Muscle Activities

Due to the constantly increasing demand for the automatic localization of landmarks in hand gesture recognition, there is a need for a more sustainable, intelligent, and reliable system for hand gesture recognition. The main purpose of this study was to develop an accurate hand gesture recognition system that is capable of error-free auto-landmark localization of any gesture dateable in an RGB image. In this paper, we propose a system based on landmark extraction from RGB images regardless of the environment. The extraction of gestures is performed via two methods, namely, fused and directional image methods. The fused method produced greater extracted gesture recognition accuracy. In the proposed system, hand gesture recognition (HGR) is done via several different methods, namely, (1) HGR via point-based features, which consist of (i) distance features, (ii) angular features, and (iii) geometric features; (2) HGR via full hand features, which are composed of (i) SONG mesh geometry and (ii) active model. To optimize these features, we applied gray wolf optimization. After optimization, a reweighted genetic algorithm was used for classification and gesture recognition. Experimentation was performed on five challenging datasets: Sign Word, Dexter1, Dexter + Object, STB, and NYU. Experimental results proved that auto landmark localization with the proposed feature extraction technique is an efficient approach towards developing a robust HGR system. The classification results of the reweighted genetic algorithm were compared with Artificial Neural Network (ANN) and decision tree. The developed system plays a significant role in healthcare muscle exercise.

Entropy in Image Analysis II

Image analysis is a fundamental task for any application where extracting information from images is required [...].

Automatic Recognition of Human Interaction via Hybrid Descriptors and Maximum Entropy Markov Model Using Depth Sensors

Automatic identification of human interaction is a challenging task especially in dynamic environments with cluttered backgrounds from video sequences. Advancements in computer vision sensor technologies provide powerful effects in human interaction recognition (HIR) during routine daily life. In this paper, we propose a novel features extraction method which incorporates robust entropy optimization and an efficient Maximum Entropy Markov Model (MEMM) for HIR via multiple vision sensors. The main objectives of proposed methodology are: (1) to propose a hybrid of four novel features-i.e., spatio-temporal features, energy-based features, shape based angular and geometric features-and a motion-orthogonal histogram of oriented gradient (MO-HOG); (2) to encode hybrid feature descriptors using a codebook, a Gaussian mixture model (GMM) and fisher encoding; (3) to optimize the encoded feature using a cross entropy optimization function; (4) to apply a MEMM classification algorithm to examine empirical expectations and highest entropy, which measure pattern variances to achieve outperformed HIR accuracy results. Our system is tested over three well-known datasets: SBU Kinect interaction; UoL 3D social activity; UT-interaction datasets. Through wide experimentations, the proposed features extraction algorithm, along with cross entropy optimization, has achieved the average accuracy rate of 91.25% with SBU, 90.4% with UoL and 87.4% with UT-Interaction datasets. The proposed HIR system will be applicable to a wide variety of man-machine interfaces, such as public-place surveillance, future medical applications, virtual reality, fitness exercises and 3D interactive gaming.