This paper concerns the problem of facial landmark detection. We provide a unique new analysis features produced at intermediate layers convolutional neural network (CNN) trained to regress coordinates. shows that while being processed by CNN, face images can be partitioned in an unsupervised manner into subsets containing faces similar poses (i.e., 3D views) and properties (e.g., presence or absence eye-wear). Based on this finding, we describe novel CNN architecture, specialized...

We propose a method designed to push the frontiers of unconstrained face recognition in wild with an emphasis on extreme out-of-plane pose variations. Existing methods either expect single model learn invariance by training massive amounts data or else normalize images aligning faces frontal pose. Contrary these, our is explicitly tackle Our proposed Pose-Aware Models (PAM) process image using several pose-specific, deep convolutional neural networks (CNN). 3D rendering used synthesize...

Traditionally, vegetable and fruit production has relied on empirical ambiguous decisions made by human farmers. To overcome this uncertainty in agriculture, smart farm robots have been widely studied recent years. However, measuring growth information with remains a challenge because of the similarity appearance target plant those around it. In study, we propose robot that accurately measures based object detection, image fusion, data augmentation fused images. The proposed uses an...

The greenhouse industry achieves stable agricultural production worldwide. Various information and communication technology techniques to model control the environment have been applied as data from environmental sensors actuators in greenhouses are monitored real time. current study designed data-based, deep learning models for evapotranspiration (ET) humidity tomato greenhouses. Using time-series applying long short-term memory (LSTM) modeling, an ET prediction was developed validated...

This paper presents a calibration method for robot Head-Eye system which can be used in humanoid vision system. estimates the transformation between coordinate and camera coordinate. The procedure uses visual measurements kinematic information as inputs of nonlinear optimization. For this method, an objective function optimization, that Minimum Variance is defined. very simple intuitive. Besides, result considerably precise robust against noisy environment. performance compared with earlier...

We present a novel convolutional neural network (CNN) design for facial landmark coordinate regression. examine the intermediate features of standard CNN trained detection and show that extracted from later, more specialized layers capture rough locations. This provides natural means applying differential treatment midway through network, tweaking processing based on alignment. The resulting Tweaked model (TCNN) harnesses robustness CNNs detection, in an appearance-sensitive manner without...

Soft sensors are attracting significant attention in human–machine interaction due to their high flexibility and adaptability. However, estimating motion state from these is difficult nonlinearity noise. In this paper, we propose a deep learning network for smart glove system predict the moving of piezoelectric soft sensor. We implemented using Long-Short Term Memory (LSTM) units demonstrated its performance real-time based on two experiments. The sensor’s was estimated joint angles were...

Recent outbreaks and the worldwide spread of COVID-19 have challenged mankind with unprecedented difficulties. The introduction autonomous disinfection robots appears to be indispensable as consistent sterilization is in desperate demand under limited manpower. In this study, we developed an navigation robot capable recognizing objects locations a high probability contamination providing quantified effects. order quantify 99.9% effect various bacterial strains, representative contaminants...

A framework to generate a human-like arm motion of humanoid robot using an Evolutionary Algorithm (EA)-based imitation learning is proposed. The consists two processes, human motions and real-time generating the database evolved in process. builds for that initially converted from capture data then genetic operator based on Principal Component Analysis (PCA) evolutionary algorithm. This evolution process also considers minimizing joint torques robot. used robot's real-time, which look like...

This paper describes the development of network based humanoids to provide services in home environment. For successful service more intelligent and varied environments, various robot sub-systems need be coordinated effectively. Thus, this also introduces a framework which makes robot-human interaction while executing tasks by means robots. Using task script, an operator can easily describe regulate actions is performing task. Furthermore, control system must operate efficiently ensure...

In recent years face recognition has made extraordinary leaps, yet unconstrained video-based identification in the wild remains an open and interesting problem. Videos, unlike still-images, offer a myriad of data for modeling, sampling, recognition, but, on other hand, contain low-quality frames motion blur. A key component is way which faces are associated through video sequence before being used recognition. this paper, we present method taking advantage body association (FBA). To track...

Cognitive architectures provide infrastructure for modeling human cognition. Recent studies reveal that these are useful control mechanisms a variety of robots. Previously, we showed one such architecture, ICARUS, can successfully humanoid robot Blocks World tasks in simulated environment. In the current work, extend application to real world and use architecture perform similar with Mahru-Z platform. As commonly expected, encountered many challenges system integration, vision-based...

Latest humanoid robots mimic the human body in realistic ways, but their control mechanisms fall short of mind. More often than not, simply have a fixed set routine codes that govern behavior, and even ones with some learning capabilities are not human-like. Since cognitive architectures aim for general intelligence provide infrastructure modeling cognition, we can benefit from this line research by adapting such as our robots. Recent studies revealed strategy is, fact, very useful....

A social robot interacts with humans based on intelligence, for which related applications are being developed across diverse fields to be increasingly integrated in modern society. In this regard, intelligence and interaction the keywords of a robot. Social refers ability control interactions or thoughts feelings relationships other people; primal empathy, is empathize by perceiving emotional signals, among components was applied study. We proposed that empathic can improved if create...

A key to successful face recognition is accurate and reliable alignment using automatically-detected facial landmarks. Given this strong dependency between landmark detection, robust requires knowledge of when the detection algorithm succeeds it fails. Facial confidence represents measure success. In paper, we propose two methods confidence: local based on predictors each landmark, global a 3D rendered model. score fusion approach also introduced integrate these confidences effectively. We...

Based on our previous work that integrated a cognitive architecture, Icarus, and humanoid robot, Mahru, we extend use of the architecture to higher-level capabilities problem solving learning. During this process, found need address known limitation ICARUS's solver. An extension is made allow handling underspecified goals in system, demonstrate new ability blocks world task involves alphabet writing with several identical blocks.

Estimation of 6D object poses is a key issue in robotic grasping tasks. Recently, many high-performance learning-based methods have been introduced using robust deep learning techniques; however, applying these to real robot environments requires ground truth pose annotations for training. To address this problem, we propose template matching-based particle filter approach estimation; the proposed method does not require poses. Although approaches can stochastically avoid local optima, they...

The COVID-19 pandemic has caused a global public health crisis, leading to increased costs for operating essential quarantine facilities and risks of infection medical staff. To address these issues, we have developed non-contact delivery robot, UTD-pro, that can deliver food supplies patients in an isolated residential treatment center. In case emergency situations such as communication failure with the control system or malfunction specific modules, noncontact robot should be able perform...

For a service robot to assist humans, it should interact with objects of varying sizes and shapes existing in an indoor environment. 3D object detection must be preceded achieve this goal since provides the ability perceive visual information. Most methods are anchor-based predict bounding box close ground truth among multiple candidates. However, is complex compute Intersection over Union (IoU) Non-Maximum Suppression (NMS) per each anchor box. Therefore, we propose keypoint-based monocular...