Abstract

El moviment és un atribut perceptiu del cervell humà molt important. La percepció visual que fa el cervell del moviment és el procés d’inferir la velocitat i direcció dels elements d’un escenari mitjançant entrades visuals. Anàlogament, la visió per computador s’assisteix mitjançant informació del moviment de l’escena. En visió per computador, la detecció de moviment és útil per a resoldre problemes com per exemple segmentació, estimació de la profunditat, estimació de l’estructura a partir del moviment, compressió de dades o navegació entre d’altres. Aquests problemes són comuns a diferents aplicacions, com ara vídeo vigilància, navegació de robots i sistemes avançats d’assistència a la conducció (Advanced Driver Assistance Systems, ADAS). Una de les tècniques més utilitzades per a detectar moviment, és el càlcul d’optical flow. El treball tractat en aquesta tesi pretén que les formulacions d’optical flow siguin més apropiades als requeriments i condicions dels escenaris de conducció. En aquest context, es proposa una nova representació de l’espai-variant anomenada representació reverse log-polar, i es demostra que, quan s’utilitza per a ADAS, té un rendiment millor que la tradicional representació log-polar. La representació espai-variant redueix la quantitat de dades necessàries que han de ser processades. Una altra contribució important està relacionada amb l’anàlisi de la influència de les característiques específiques d’escenaris de conducció per a la precisió de l’optical flow. S’han considerat característiques tals com la velocitat del vehicle i la textura de la carretera. D’aquest estudi s’infereix que, el pes del terme de regularització s’ha d’adaptar segons una mesura d’error i per a diferents velocitats i textures de la carretera. També es mostra que la representació polar d’optical flow funciona molt millor per a escenaris de conducció on el moviment principal són translacions. Degut als requeriments d’aquest estudi, i per la manca de bases de dades es presenta una nova base de dades sintètica que conté: i) seqüències amb diferents velocitats i textures en un escenari urbà; ii) seqüències amb moviments complexos de la càmera col·locada al vehicle; i iii) seqüències amb altres vehicles en moviment dintre la mateixa escena. L’optical flow corresponent a cada seqüència s’obté mitjançant la tècnica de ray-tracing. A més a més, es presenten algunes aplicacions per a optical flow en escenaris ADAS. Per començar, proposem una tècnica robusta basada en RANSAC per estimar la línia de l’horitzó. Després, presentem una estimació de l’egomotion per a comparar la representació espai-variant proposada amb les representacions clàssiques. Com a contribució final, es proposa una modificació del terme de regularització que millora notablement els resultats per a aplicacions d’ADAS. Aquesta adaptació s’avalua mitjançant tècniques d’optical flow d’última generació. Els experiments realitzats amb una base de dades pública (KITTI) validen els avantatges d’utilitzar la modificació proposada. La percepción del movimiento es uno de los más importantes atributos del cerebro humano. La percepción visual del movimiento consiste en inferir velocidad y dirección de los elementos móviles que interactúan en una escena, mediante la interpretación de diferentes entradas visuales. Análogamente, la visión por computador hace uso de la información del movimiento en la escena. La detección de movimiento en visión por computador es útil para resolver problemas tales como: segmentación, estimación de profundidad, compresión, navegación, entre otros. Estos problemas son comunes en distintas aplicaciones, por ejemplo: video vigilancia, navegación de robots y sistemas avanzados de asistencia a la conducción (ADAS). Una de las técnicas más utilizadas para detectar movimiento es la estimación del flujo óptico. El trabajo abordado en esta tesis busca formulaciones del flujo óptico más adecuadas a las necesidades y condiciones de los escenarios de conducción. En este contexto, se propuso una novedosa representación del espacio, llamada representación inversa log-polar, la cual se demuestra que tiene un desempeño mejor que la tradicional representación logpolar para aplicaciones ADAS. Las representaciones de espacio-variante reducen la cantidad de datos a ser procesados. Otra contribución importante está relacionada con el análisis de la influencia de las características específicas de los escenarios de conducción en la precisión del flujo óptico estimado. Características tales como la velocidad del vehículo y la textura de la carretera son consideradas en el estudio. De este estudio, se infiere que el peso del término de regularización tiene que ser adaptado de acuerdo con la medida de error requerida y para diferentes velocidades y texturas de la carretera. También se concluye que la representación polar del flujo óptico es la más apropiada en escenarios de conducción, donde el movimiento predominante es la translación. Debido a las exigencias de tal estudio, y por falta de las bases de datos necesarias, se presenta un nuevo conjunto de datos sintéticos el cual contiene: i) secuencias de diferentes velocidades y texturas en un escenario urbano; ii) secuencias con movimientos complejos de la cámara dispuesta en el vehículo; y iii) secuencias con otros vehículos en movimiento en la escena. El flujo óptico correspondiente a cada secuencia es obtenido mediante una técnica de ray-tracing. Adicionalmente, se presentan algunas aplicaciones de flujo óptico en ADAS. Primeramente se propone una técnica robusta basada en RANSAC para estimar la línea de horizonte. Seguidamente se presenta una estimación del egomotion para comparar la representación de espacio-variante propuesta con los esquemas clásicos. Como contribución final, se propone una modificación en el término de regularización que mejora notablemente los resultados en las aplicaciones ADAS. Los resultados experimentales en una base de datos pública (KITTI) validan las ventajas de la utilización de la modificación propuesta. Motion perception is one of the most important attributes of the human brain. Visual motion perception consists in inferring speed and direction of elements in a scene based on visual inputs. Analogously, computer vision is assisted by motion cues in the scene. Motion detection in computer vision is useful in solving problems such as segmentation, depth from motion, structure from motion, compression, navigation and many others. These problems are common in several applications, for instance, video surveillance, robot navigation and advanced driver assistance systems (ADAS). One of the most widely used techniques for motion detection is the optical flow estimation. The work in this thesis attempts to make optical flow suitable for the requirements and conditions of driving scenarios. In this context, a novel space-variant representation called reverse log-polar representation is proposed that is shown to be better than the traditional log-polar space-variant representation for ADAS. The space-variant representations reduce the amount of data to be processed. Another major contribution in this research is related to the analysis of the influence of specific characteristics from driving scenarios on the optical flow accuracy. Characteristics such as vehicle speed and road texture are considered in the aforementioned analysis. From this study, it is inferred that the regularization weight has to be adapted according to the required error measure and for different speeds and road textures. It is also shown that polar represented optical flow suits driving scenarios where predominant motion is translation. Due to the requirements of such a study and by the lack of needed datasets a new synthetic dataset is presented; it contains: i) sequences of different speeds and road textures in an urban scenario; ii) sequences with complex motion of an on-board camera; and iii) sequences with additional moving vehicles in the scene. The ground-truth optical flow is generated by the ray-tracing technique. Further, few applications of optical flow in ADAS are shown. Firstly, a robust RANSAC based technique to estimate horizon line is proposed. Then, an egomotion estimation is presented to compare the proposed space-variant representation with the classical one. As a final contribution, a modification in the regularization term is proposed that notably improves the results in the ADAS applications. This adaptation is evaluated using a state of the art optical flow technique. The experiments on a public dataset (KITTI) validate the advantages of using the proposed modification.

Abstract

Inland navigation networks are large-scale systems that can be described by using the nonlinear Saint-Venant partial differential equations. However, as there is no analytical solution for them, simplified models are used instead for modeling purposes. This work addresses the modeling of two sub-reach systems by means of the well-known Integrator Delay Zero model. Two main scenarios are considered: in the first one, the two partial models are independently computed one from each other; the second one uses previous knowledge of the whole two sub-reach system in order to ensure the flow consistency along the system. The application of these two methodologies to a part of the navigation network in the north of France serves as the case study for this work. Peer Reviewed

Abstract

Cities have become the most common living environment for humans. With this rising urbanization, urban design has become vital for these growing cities. While measuring objective data like traffic congestion or air quality is important, it does not tell the whole story of how people live in the cities or how cities should be developed to make them more livable. In future for a true smart city a more humane component is needed to understand how the population of cities actually interact with and feel about their surroundings. Surveys are a great and a necessary tool for this and they are already being used in the design process. However, they require effort and and a lot of silent information can be missed. The surveying process also doesn't happen in real time. We suggest that social media data could be used to gather more information about human- environment interaction in cities and compliment the surveys. We show a working prototype of a tool that creates an emotional map of a city by mining social media data for sentiments and heatmapping them. This kind of method could prove to be an useful tool for urban designers, who could take advantage of the visual intuition of humans and see instantly where and how emotional hotspots arise. It could also be of interest for emotion researchers, who could get data on what it really means to be happy for a human being - for example eating an ice cream at the beach - instead of only linking conceptual words (such as happy) to external stimuli (such as smiling). Peer reviewed

Abstract

new asynchronous pipeline controller based on early acknowledgement protocol is proposed. The early acknowledgement protocol indicates acknowledgment by the falling edge of the acknowledgment signal, and thus, it can hide the overhead caused by a resetting phase of the handshake cycle. Advantage of employing the early acknowledgement protocol in a pipeline controller is demonstrated by comparing its performance with a very high-speed 2-phase pipeline controller and an ordinary 4-phase controller. The simulation results obtained in our experiment show that in a linear pipeline case with processing elements in-between the stages, our controller has performance comparable to the 2-phase controller, and outperforms the ordinarily 4-phase controller. Moreover, we have obtained interesting results in the case of a nonlinear pipeline with a Conditional Branch operation. In this case, a 2-phase controller needs to have a little complicated mechanism to handle the 2-phase operation for selecting a path, while our controller is trivial in construction similar to the ordinary 4-phase protocol controller. Due to such simplicity of the implementation, our controller has a slightly better performance compared to the 2-phase controller when each stage has a processing element.

Abstract

xtabstractWe investigate the impact of random deviations in planned arrival times on user equilibrium in an extension of Vickrey’s celebrated bottleneck model. The model is motivated by the fact that in real life, users can not exactly plan the time at which they depart from home, nor the delay they experience before they join the congestion bottleneck under investigation. We show that the arrival density advocated by the Nash equilibrium in Vickrey’s model, is not a user equilibrium in the model with random uncertainty. We then investigate the existence of a user equilibrium for the latter and show that in general such an equilibrium can neither be a pure Nash equilibrium, nor a mixed equilibrium with a continuous density. Our results imply that when random distortions influence user decisions, the dynamics of standard bottleneck models are inadequate to describe such more complex situations. We illustrate with numerical analysis how the mechanics of a bottleneck with delayed arrivals are unstable for any continuous arrival strategy, thus shedding more light on the non-existence result.

Abstract

International audience; When displaying thousands of aircraft trajectories on a screen, the visualization is spoiled by a tangle of trails. The visual analysis is therefore difficult, especially if a specific class of trajectories in an erroneous dataset has to be studied. We designed FromDaDy, a trajectory visualization tool that tackles the difficulties of exploring the visualization of multiple trails. This multidimensional data exploration is based on scatterplots, brushing, pick and drop, juxtaposed views and rapid visual design. Users can organize the workspace composed of multiple juxtaposed views. They can define the visual configuration of the views by connecting data dimensions from the dataset to Bertin's visual variables. They can then brush trajectories, and with a pick and drop operation they can spread the brushed information across views. They can then repeat these interactions, until they extract a set of relevant data, thus formulating complex queries. Through two real-world scenarios, we show how FromDaDy supports iterative queries and the extraction of trajectories in a dataset that contains up to 5 million data.

Abstract

Event cameras are bio-inspired vision sensors that output pixel-level brightness changes instead of standard intensity frames. These cameras do not suffer from motion blur and have a very high dynamic range, which enables them to provide reliable visual information during high-speed motions or in scenes characterized by high dynamic range. These features, along with a very low power consumption, make event cameras an ideal complement to standard cameras for VR/AR and video game applications. With these applications in mind, this paper tackles the problem of accurate, low-latency tracking of an event camera from an existing photometric depth map (i.e., intensity plus depth information) built via classic dense reconstruction pipelines. Our approach tracks the 6-DOF pose of the event camera upon the arrival of each event, thus virtually eliminating latency. We successfully evaluate the method in both indoor and outdoor scenes and show that---because of the technological advantages of the event camera---our pipeline works in scenes characterized by high-speed motion, which are still unaccessible to standard cameras. Comment: 12 pages, 13 figures. 2 tables. (in press)

Abstract

Spatial time series is a common type of data dealt with in many domains, such as economic statistics and environmental science. There have been many studies focusing on finding and analyzing various kinds of events in time series; the term ‘event’ refers to significant changes or occurrences of particular patterns formed by consecutive attribute values. We focus on a further step in event analysis: discover temporal relationship patterns between event locations, i.e., repeated cases when there is a specific temporal relationship (same time, before, or after) between events occurring at two locations. This can provide important clues for understanding the formation and spreading mechanisms of events and interdependencies among spatial locations. We propose a visual exploration framework COPE (Co-Occurrence Pattern Exploration), which allows users to extract events of interest from data and detect various co-occurrence patterns among them. Case studies and expert reviews were conducted to verify the effectiveness and scalability of COPE using two real-world datasets.

Abstract

In this work, we present an interactive system for visual analysis of urban traffic congestion based on GPS trajectories. For these trajectories we develop strategies to extract and derive traffic jam information. After cleaning the trajectories, they are matched to a road network. Subsequently, traffic speed on each road segment is computed and traffic jam events are automatically detected. Spatially and temporally related events are concatenated in, so-called, traffic jam propagation graphs. These graphs form a high-level description of a traffic jam and its propagation in time and space. Our system provides multiple views for visually exploring and analyzing the traffic condition of a large city as a whole, on the level of propagation graphs, and on road segment level. Case studies with 24 days of taxi GPS trajectories collected in Beijing demonstrate the effectiveness of our system.