Deep learning has shown impressive performances in many problems of artificial intelligence including face analysis which has been a challenging task in computer vision for decades. Face analysis has various application areas such as video surveillance, human-computer interaction and driver fatigue detection. In this talk, I will give a brief overview of deep learning approaches in face analysis. After detecting the face regions in an input image the first step is to pre-process those regions for the subsequent stages. Although the pre-processing operations differ based on the application of analysis I will be focusing on the deep learning methods developed for face alignment, pose estimation, face formalization and face super-resolution problems. Other subjects in the talk are related to face attribute estimation, face expression and emotion analysis and face recognition. Face attributes include age, gender and other attributes such as glass/no glass and beard/no beard. In facial expression/emotion analysis I will discuss the methods in recognition of prototypical facial expressions, i.e. anger, disgust, fear, happiness, sadness and surprise in addition to facial action units. The face recognition part of the talk covers both the identification and verification problems. The talk will focus on state-of-the-art deep learning studies in each face analysis problem after grouping the main approaches. Finally I conclude my talk with the challenges and future works on face analysis.

Due largely to the growing emphasis in academic research on industrial robots and their applications, it is oft en required by researchers to understand and examine the kinematic aspect of such robots. Obtaining both forward and inverse kinematic models of a given industrial robot could be a tedious and intricate task. Th erefore, this paper presents a detailed kinematic analysis of a 6-axis industrial robot that commonly found in present-day industry and research laboratories. Th e proposed kinematic solutions have been validated against the simulation soft ware provided by robot’s supplier and an error analysis has been done to ensure the accuracy. Th ere are particular models of robots which are well discussed in the fi eld of robotic research and also within the domain of kinematic analysis. Th e Puma 560 by Unimation, also known as “white rat” of robotics is one of such robots which catalyzed robotic research for decades, and well examined in textbooks and research articles. However, the eminence of such robots is gradually being replaced by modern 6-axis industrial robots. Perhaps due to its size, prize, availability or consideration of health and safety aspects, the Fanuc LR Mate 200iD is becoming increasingly popular in the industry as well as in the research laboratories. Alternatively, there are a limited number of research articles which examine the kinematics of present day industrial robots. Further, the majority of those articles barely discusses the results and lacks the scientifi c validation of proposed solutions. Th erefore, the main focus of this article was to generate and validate the both forward and inverse kinematic models of a popular and modern industrial robot – the Fanuc LR Mate 200iD.

Color is a powerful form of communication among human beings. Sociable robots that live and coexist with humans must also learn colors from the society it lives. A lot of research has been performed to enable computer, as the brain of a robot, to learn colors. Most of them rely on modeling of human color perception and mathematical complexities. Diff erently, this work targets on developing the capability of the computer to use machine learning approaches to learn the colors through human interaction. Th e diff erent colors which is being detected in the camera is processed by using image processing tool OpenCV and the most dominant color of the picture is identifi ed and displayed in the system. Th e user can now teach the computer the difference between the appropriate colors using the RGB values. Th erefore, although at the beginning, the computer does not know any colors, eventually through interaction, it learns numerous colors which will indicate the shared color learning with humans in the society. Aft er teaching the computer a number of times, it is able to classify the colors by matching with RGB values for that particular color from the database. If the color does not exist, the computer identifi es the closest possible color using the unsupervised machine learning technique k-means clustering. Aft er learning colors from the society, the developed algorithm is implemented in the NTU Singaboat, which is an Unmanned Surface Vehicle (USV) built for competing in the Maritime RobotX Challenge.

This article refers to a phenomenological model of a batch reactor. Using an energy balance model it was possible to predict the heating and cooling behavior inside the reactor and how the temperature varies with the applied voltage. Th is work will be used for later emphasis on polymer syntheses, especially the hydrolytic synthesis of caprolactam, in order to obtain a product with a higher value-added market, nylon 6. Given the diffi culty of temperature control when operating with nylon 6, silicone oil (nylon-like characteristics) was used for system testing and data collection. In this way, the system boils down to a heating tank, with heating via electrical resistance. Th us, the focus of the present study will be on modeling the reactor using silicone oil. Th e study will be done obtaining dynamic measurements of temperature in order to be able to present a phenomenological model of the reactor. For the validation of the model, we used data collected in the plant from a steptype test. In this way, this text aims to develop the phenomenological model of the system in order to better understand the dynamics of the heating and thus enable future control studies. Th e model represented a satisfactory behavior of the reactor in question, presenting an average relative error of 4.3%.