Virtual crowd rendering have been used in film industry with offine rendering methods for a long time. But its existence in interactive real-time applications such as video games is not so common due to the limited rendering power of current graphics hardware. This thesis describes a novel method to improve shadow mapping performance of a crowded scene by taking into account the screen space visibility of the casted shadow of a crowd instance when rendering the shadow maps. A grid-based visibility mask creation method is proposed which is irrelevant to scene complexity. This improves the rendering performance especially when there are many occluded instances of the crowd which is a common scenario in urban environments and accelerates the usage of crowds in real time applications, such as games. We compute visibility of all agents in a crowd in parallel on the graphics processing unit(GPU) without having a requirement of a stencil buer or light direction dependent shadow mask. Technique also improves the view space rendering time by reducing the visibility check cost of the agents that are located on the invisible areas of the scene. The methodology introduced in this thesis gets more effective in each shadow map rendering pass by re-using the same visibility mask for shadow caster culling and enables many local lights with shadows. We also give a brief information about the state of the art of crowd rendering and shadowing, explaining how suitable the method with the implementations of different shadow mapping approaches. The technique is very well compatible with the modern crowd rendering techniques such as skinned instancing, dynamic level of detail(LOD) determination and GPU-based simulation.

Pedestrian crowd simulation explores crowd behaviors in virtual environments. It is extensively studied in many areas, such as safety and civil engineering, transportation, social science, entertainment industry and so on. As a common phenomenon in pedestrian crowds, grouping can play important roles in crowd behaviors. To achieve more realistic simulations, it is important to support group modeling in crowd behaviors. Nevertheless, group modeling is still an open and challenging problem. The influence of groups on the dynamics of crowd movement has not been incorporated into most existing crowd models because of the complexity nature of social groups. This research develops a framework for group modeling in agent-based pedestrian crowd simulations. The framework includes multiple layers that support a systematic approach for modeling social groups in pedestrian crowd simulations. These layers include a simulation engine layer that provides efficient simulation engines to simulate the crowd model; a behavior-based agent modeling layers that supports developing agent models using the developed BehaviorSim simulation software; a group modeling layer that provides a well-defined way to model inter-group relationships and intra-group connections among pedestrian agents in a crowd; and finally a context modeling layer that allows users to incorporate various social and psychological models into the study of social groups in pedestrian crowd. Each layer utilizes the layer below it to fulfill its functionality, and together these layers provide an integrated framework for supporting group modeling in pedestrian crowd simulations. To our knowledge this work is the first one to focus on a systematic group modeling approach for pedestrian crowd simulations. This systematic modeling approach allows users to create social group simulation models in a well-defined way for studying the effect of social and psychological factors on crowd’s grouping behavior. To demonstrate the capability of the group modeling framework, we developed an application of dynamic grouping for pedestrian crowd simulations.

Has it ever occurred to you to wonder why a soldier would sacrifice his life by jumping on a bomb to save the rest of his brigade? Or why an individual in a gang might display respectable behavior when alone but swear and vandalize when in the group? The phenomenon of people getting pulled into crowds and adopting the group’s mentalities and behaviors has been recognized but not fully researched. However, it has been recorded in early literature and research that it is human nature to want to fit into a group, for example in Abraham Maslow’s (1943) paper, A Theory of Human Motivation, in which he proposed that the hierarchy of human needs includes a stage that emphasized an individual’s need to feel a sense of belonging.

Recently, the surveillance of human activities has drawn a lot of attention from the research community and the camera based surveillance is being tried with the aid of computers. Cameras are being used extensively for surveilling human activities; however, placing cameras and transmitting visual data is not the end of a surveillance system. Surveillance needs to detect abnormal or unwanted activities. Such abnormal activities are very infrequent as compared to regular activities. At present, surveillance is done manually, where the job of operators is to watch a set of surveillance video screens to discover an abnormal event. This is expensive and prone to error. The limitation of these surveillance systems can be effectively removed if an automated anomaly detection system is designed. With powerful computers, computer vision is being seen as a panacea for surveillance. A computer vision aided anomaly detection system will enable the selection of those video frames which contain an anomaly, and only those selected frames will be used for manual verifications. A panic is a type of anomaly in a human crowd, which appears when a group of people start to move faster than the usual speed. Such situations can arise due to a fearsome activity near a crowd such as fight, robbery, riot, etc. A variety of computer vision based algorithms have been developed to detect panic in human crowds, however, most of the proposed algorithms are computationally expensive and hence too slow to be real-time. Dictionary learning is a robust tool to model a behaviour in terms of the linear combination of dictionary elements. A few panic detection algorithms have shown high accuracy using the dictionary learning method; however, the dictionary learning approach is computationally expensive. Orthogonal matching pursuit (OMP) is an inexpensive way to model a behaviour using dictionary elements and in this research OMP is used to design a panic detection algorithm. The proposed algorithm has been tested on two datasets and results are found to be comparable to state-of-the-art algorithms.

群體運動在現今的電玩、動畫或電影中，有十分重要的應用；透過群體性的運動，可以表現出故事背景設定的張力。在群體運動的模擬中，除了個體本身的運動行為模擬外，重要的是如何呈現出群體運動的整體效果。過去文獻中多數的群體運動模擬系統，要能在群體運動中呈現出特定形狀的效果，多需花費大量的時間反覆調整個體的模擬結果；個體本身的運動行為模型，則多採用虛擬力場的方式，被動的影響個體的運動，較缺乏直觀設定行為模型的方式。本論文的目標是建立一套人群模擬系統，此系統包括兩個部份：第一個部份是使用者可根據個人偏好設定群體運動理想中的外觀形狀，使此系統在模擬前能根據所輸入的環境資訊，利用運動計畫的方式，自動產生形體形變的路徑，以做為維持群體外形的參考目標。第二部份則是改進人群模擬時個體與群體的行為模型。我們利用模糊數學的特性，來表示行為模型以語意表達時的不確定性，使個體的行為能表現出貼近使用者所需求的結果。我們提出了三種類型的模糊行為模型與對應的原生動作，用以表現個體與群體的運動行為。根據我們實做出來的系統及實驗顯示，透過這樣的系統，我們可以利用程序化的方式為電腦動畫師產生具有特定群體外觀的群體模擬，減少其在製作相關動畫所需要的時間與技術成本，同時也提供了直觀的方式建立人群的行為模型，增加行為的豐富性。 / The effects of crowd behavior are becoming indispensable in computer games and computer animation. In crowd simulation, beside the issue of simulating the motion for individual agents, the more important one is how to simulate a specific behavior of a crowd based on the motion of individuals. In order for a crowd to conform to a specific shape, most simulation systems reported in the literature require the users to spend a great amount of time in tuning the behavior parameters of each individual, governed by virtual forces computed according to inter-agent relations. In this thesis, we aim to build a system of crowd simulation consisting two parts: a path planner for a flexible shape and a motion controller with fuzzy logic. The path planner can search for a feasible path for a region of flexible shape allowing the user to set his preference on the shape of the crowd. The local motion controller for each agent is based on fuzzy logic rules that can be used to present the uncertainty of linguistic behavior models. We have proposed three types of fuzzy behavior models and their corresponding primitive actions. Our experiments show that, with this simulation system, we allow a computer animator to use an intuitive way to create specific appearance and richer.

本文嘗試透過社群媒體微博進行憂鬱書寫識別，主要期望回答兩方面的問題：（一）中國憂鬱人群之社群媒體書寫特質為何？（二）如何透過該書寫特質識別更多的憂鬱文本？ 透過對十位已確認之憂鬱症患者之微博關係圈進行滾雪球，發現 127憂鬱傾向者，共爬取憂鬱傾向者之微博文本20748則，作為文本分析之數據集，並運用內容分析、質化分析、詞頻分析及詞語共現等多種方法分析文本。 分析結果顯示：(1)透過對文本進行語調、情緒、主題及憂鬱程度的編碼後，我們發現憂鬱傾向者在微博之書寫含62%的負面語調及25.1%的憂鬱文本，其中，負面及憂鬱程度較高的書寫主題是「自我」、「親情」、「自殺」及「睡眠障礙」。(2)深入對「自我」及「親情」憂鬱書寫的質化分析後，發現他們不同於一般人的心理特質，其中，「自我厭惡」及「不被理解」是他們心中最難以釋懷的角落。(3)由於「自殺」、「睡眠障礙」屬於憂鬱人群特徵，經過分析發現透過主題關聯詞的共現詞組有助於辨識憂鬱人群，其中，「睡眠障礙」共現詞的憂鬱文本辨識度達74%，「自殺」共現詞的憂鬱文本辨識度達34%，未來透過機器的方式，可進一步優化該方法，提升憂鬱文本的辨識度。 / This research aims to answer the following questions:(1)What are the characteristics of micro-blog writing by the depressed tendency people? (2)How to identify the text in social media? Ten Wei-bo users with identified depressed tendency were chosen as starting points of snow-ball searching, and 127 users were located. A total of 20748 messages from this group of the users was collected as the dataset. Multiple methods were applied to analyze the texts: content analysis, qualitative text analysis, word frequency analysis and word co-occurrence. The result indicated that: (1)Through the coding of the text tone, mood, theme and degree of depression, we find out that in micro blog writing, the depressive tendency uses 62% of the negative tone and 25.1% of the blue text. Among them, higher negative and degree of depression of writing subjects are "self", "family", "suicide" and "sleep disorder". (2)Through deep qualitative analysis of "self" and "affection" depressed writing, the "self loathing" and "don't understand" in their mind are the most unforgettable. (3)Because the depressed people have the features of "suicide" and "sleep disorder", through the analysis, we find that through theme related words, it is helpful in the identification of the depression text. Among them, the "sleep disorders" co-occurrence words depressed text identification is up to 74%, and "suicide" co-occurrence words depressed text identification degree is 34 %.In the future, through the computer, we can further optimize the method, and enhance the degree of identification of depression text.

Recently, the surveillance of human activities has drawn a lot of attention from the research community and the camera based surveillance is being tried with the aid of computers. Cameras are being used extensively for surveilling human activities; however, placing cameras and transmitting visual data is not the end of a surveillance system. Surveillance needs to detect abnormal or unwanted activities. Such abnormal activities are very infrequent as compared to regular activities. At present, surveillance is done manually, where the job of operators is to watch a set of surveillance video screens to discover an abnormal event. This is expensive and prone to error. The limitation of these surveillance systems can be effectively removed if an automated anomaly detection system is designed. With powerful computers, computer vision is being seen as a panacea for surveillance. A computer vision aided anomaly detection system will enable the selection of those video frames which contain an anomaly, and only those selected frames will be used for manual verifications. A panic is a type of anomaly in a human crowd, which appears when a group of people start to move faster than the usual speed. Such situations can arise due to a fearsome activity near a crowd such as fight, robbery, riot, etc. A variety of computer vision based algorithms have been developed to detect panic in human crowds, however, most of the proposed algorithms are computationally expensive and hence too slow to be real-time. Dictionary learning is a robust tool to model a behaviour in terms of the linear combination of dictionary elements. A few panic detection algorithms have shown high accuracy using the dictionary learning method; however, the dictionary learning approach is computationally expensive. Orthogonal matching pursuit (OMP) is an inexpensive way to model a behaviour using dictionary elements and in this research OMP is used to design a panic detection algorithm. The proposed algorithm has been tested on two datasets and results are found to be comparable to state-of-the-art algorithms.

十八世紀工業革命的推進，使得大量生產(Mass Production)幾乎改變了各產業，為社會帶來大量的財富，隨著網際網路以及社群網站的普及化，大量生產已經無法應付日漸複雜的市場環境，而需要新的商業模式來達到突破。 隨著社群的概念興起，群眾外包的觀念也隨之廣泛應用於各種產業，而不再只限於開放原始碼的用途，逐漸被應用於T-shirt、科技業、雜誌業等等的範疇，舉凡美國Threadless.com、Innocentive等等都是應用群眾外包之成功案例。 群眾外包即是指提供平台供群眾使用，並在該平台上提供創意發想的點子，最終經由一定的表決機制發展成新產品，而非傳統商品化方式─由廠商開發製造完成新產品。 本研究探討群眾外包應用於台灣服飾業者之商品化流程，進而找出關鍵成功策略。為了有效釐清群眾外包商品化之複雜關記以及與群眾之互動細節，採用多重個案研究法，該質化的研究方法可由深度訪談產業專家，以了解發展歷程，並藉由個案廠商的角度，探討群眾外包之關鍵策略。 研究發現，群眾外包之策略是否可行有以下四個檢核點：群眾獎勵機制、仲介網路平台、評選機制以及生產與營運。首先要建立凝聚相同興趣的社群，並提供自由發揮的平台，藉由公正且有效率的評選機制選出欲生產之商品，透過有效率的生產才能順利將商品打入市場。 / As internet become more and more popular, customers become pickier because all of the information is so clear in it. Also, it makes the social network become stronger, and become the new method for enterprises to obtain ideas and market their goods. That’s how crowd sourcing has been used in many industries. This study investigates how Taiwan T-shirt enterprises to use crowd sourcing as a method to obtain more works from the crowds on the internet. With the longitudinal study of three companies to investigate processes and content of crowd sourcing strategies. Study found out that the crowd sourcing strategy can be cut into four key points: give awards to attract people with the same interest involving in, establish a web platform for people to share ideas and put their works on, establish a fair appraise mechanism, and manufacture in a efficient way.

Orientadora: Rosangela da Silva Leote / Banca: Milton Terumitsu Sogabe / Banca: Giselle Beiguelman / Resumo: Este trabalho analisa o conceito de commons digital, originado no campo da cibercultura, no contexto de ações artísticas produzidas nos últimos anos na cidade de São Paulo. Valendome principalmente das teorias de Antonio Negri, Nicolas Bourriaud, Guy Debord e Gilles Lipovetsky, parto da retomada do "comum" como valor intrínseco à sociedade informacional, sigo com sua análise nos campos da arte pública e artemídia para, enfim, classificar cinco aspectos das poéticas do comum (p. 118), a partir de três modelos de casos: BaixoCentro (2012), Pimp My Carroça (2012) e Piscina no Minhocão (2014). Por fim, apresento os primeiros resultados da ação Terrenos Apaixonantemente Objetivos, que aplica a ideia do comum à deriva situacionista, e detalho a concepção e desenvolvimento do Derivoscópio, obra integrante da ação, um aparato vestível construído com hardware e software livres. / Abstract: This work analyzes the concept of digital commons, originated in the field of cyberculture, in the context of artistic actions produced in the last years in the city of Sao Paulo, Brazil. Based primarily on the theories by Antonio Negri, Nicolas Bourriaud, Guy Debord and Gilles Lipovetsky, I start from the resumption of the "common" as an intrinsic value to the Information Society to proceed with its analysis in the fields of Public Art and New Media Art. Then, I classify five aspects of the poetics of the common (p. 118), based on three case models: BaixoCentro (2012), Pimp My Carroça (2012) and Piscina no Minhocão (2014). Lastly, I present the first results of the artwork Objective Passional Terrains, which applies the concept of the common to the situationist drift, detailing the design and development of the Driftscope, a wearable apparatus built with free hardware and software, as part of that artwork. / Mestre

In this thesis, we address several generic problems concerned with procurement of tasks from a crowd that consists of strategic workers with uncertainty in their qualities. These problems assume importance as the quality of services in a service marketplace is known to degrade when there is (unchecked) information asymmetry pertaining to quality. Moreover, crowdsourcing is increasingly being used for a wide variety of tasks these days since it offers high levels of flexibility to workers as well as employers. We seek to address the issue of quality uncertainty in crowdsourcing through mechanism design and machine learning. As the interactions in web-based crowdsourcing platform are logged, the data captured could be used to learn unknown parameters such as qualities of individual crowd workers. Further, many of these platforms invite bids by crowd workers for available tasks but the strategic workers may not bid truthfully. This warrants the use of mechanism design to induce truthful bidding. There ensues a complex interplay between machine learning and mechanism design, leading to interesting technical challenges. We resolve some generic challenges in the context of the following problems. Design of a quality eliciting mechanism with interdependent values We consider an expert sourcing problem, where a planner seeks opinions from a pool of experts. Execution of the task at an assured quality level in a cost effective manner turns out to be a mechanism design problem when the individual qualities are private information of the experts. Also, the task execution problem involves interdependent values, where truthfulness and efficiency cannot be achieved in an unrestricted setting due to an impossibility result. We propose a novel mechanism that exploits the special structure of the problem and guarantees allocative efficiency, ex-post incentive compatibility and strict budget balance for the mechanism, and ex-post individual rationality for the experts. Design of an optimal dimensional crowdsourcing auction We study the problem faced by an auctioneer who gains stochastic rewards by procuring multiple units of a service from a pool of heterogeneous strategic workers. The reward obtained depends on the inherent quality of the worker; the worker’s quality is fixed but unknown. The costs and capacities are private information of the workers. The auctioneer is required to elicit costs and capacities (making the mechanism design dimensional) and further, has to learn the qualities of the workers as well, to enable utility maximization. To solve this problem, we design a dimensional multi-armed bandit auction that maximizes the expected utility of the auctioneer subject to incentive compatibility and individual rationality while simultaneously learning the unknown qualities of the agents. Design of a multi-parameter learning mechanism for crowdsourcing We investigate the problem of allocating divisible jobs, arriving online, to workers in a crowd-sourcing platform. Each job is split into a certain number of tasks that are then allocated to workers. These tasks have to meet several constraints that depend on the worker performance. The performance of each worker in turn is characterized by several intrinsic stochastic parameters. In particular, we study a problem where each arriving job has to be completed within a deadline and each task has to be completed, honouring a lower bound on quality. The job completion time and quality of each worker are stochastic with fixed but unknown means. We propose a learning mechanism to elicit the costs truthfully while simultaneously learning the stochastic parameters. Our proposed mechanism is dominant strategy incentive compatible and ex-post individually rational with asymptotically optimal regret performance.