A Framework for Control of a Camera Head Claus
نویسندگان
چکیده
This dissertation presents an architecture for controlling an agile camera head with multiple degrees of freedom. The purpose of the architecture is to provide control of the di erent degrees of freedom in the head by applying it in a xation based scheme. This includes control of the optical degrees of freedom, an often neglected property in most current head control architectures. The basic thesis is that a tight connection between the lowest visual processes and the sensing apparatus is imperative for successful operation in dynamic environments. This has lead to the introduction of a basic system which facilitates three low level processes, which are; xation, tracking and attention selection and shifting, which can provide fully data driven responses. The basic system carries the possibility of servicing requests from higher level visual processes, whereby model driven responses in a hypothesis veri cation/rejection scheme may be implemented. The interaction between the basic system and higher level processes is solely accomplished through the attention mechanism, whereby multiple processes control of hardware is resolved. Using inspiration from the motoric system in the primate vision system an architecture based on Cyclopean representation of visual information has been designed. This is an equal eye dominance approach, which separates control of xation point into control of vergence, version and tilt angles, whereby the control of range and \position" is separated. The Cyclopean representation has been used to fuse two range cues, disparity and accommodation for controlling the two corresponding actuators, accommodation (focus) distance and vergence angle. The integration is based on fusing the cues according to their relative certainty, thus both cues in uence the two actuators. Using this algorithm we have shown the range processes able to maintain xation during monocular occlusion. An attention mechanism embedding scale space selection of interesting areas from multiple features has been designed and implemented. Experiments involving attention selection from a wide variety of scenes have been provided. The individual processes as well as the entire system have been analyzed through a number of experiments. Tracking and smooth pursuit in particular have been achieved both for very stylistic objects and more natural ones. Tracking in 3D has been achieved at 10 12Hz with objects moving up to 7 =s, with an average accuracy better than 1cm in a distance of up to 3m. Simultaneously the tracking is providing good image stabilization. The proposed architecture has proved an e cient way of organizing the low level vision system for a xation based system. This is mainly due to the separation between the speci c processing required and architecture itself. iii The thesis is in part based on the following publications: Claus Siggaard Andersen, Jan Juul S rensen, and Henrik Iskov Christensen. An analysis of three depth recovery techniques. In Proceedings of the Seventh Scandinavian Conference on Image Analysis, pages 66{77, Lab of Image Analysis, Aalborg University, Fr. Bajersvej 7, Bldg. D1, DK-9220 Aalborg East, Denmark, August 1991. Pattern Recognition Society of Denmark. Henrik Iskov Christensen, Claus Siggaard Andersen, and Erik Granum. Control of perception in dynamic vision. In S ren I. Olsen, editor, Proceedings of the First Danish Conference on Pattern Recognition and Image Analysis, pages 91{ 98, Copenhagen, Denmark, 1992. DIKU-Tryk. Claus Siggaard Andersen, Henrik Iskov Christensen, and Erik Granum. Using verged cameras in visual navigation. In Peter Johansen, editor, Proceedings of the Second Danish Conference on Pattern Recognition and Image Analysis, pages 67{77. DIKU-Tryk, 1993. Oddbj rn Bergem, Claus Siggaard Andersen, and Henrik Iskov Christensen. Using match uncertainty in the kalman lter for a sonar based positioning system. In Proceedings of the Eighth Scandinavian Conference on Image Analysis, volume 1, pages 405{410, Troms , Norway, 1993. Claus Siggaard Andersen and Henrik Iskov Christensen. Using multiple cues for controlling an agile camera head. In IAPR Workshop on Visual Behaviors, pages 97{101, Seattle, Washington, USA, June 1994. IAPR, IEEE Press. Claus Siggaard Andersen. Sensor fusion and integration a survey. Technical Report, October 1993. iv Sammenfatning Denne afhandling pr senterer en arkitektur for kontrol af et kamerahoved med b ade optiske og mekaniske frihedsgrader. Form alet med arkitekturen er at opn a kontrol af de enkelte frihedsgrader ved anvendelse som et xationsbaseret system. Dette inkluderer kontrol af de optiske frihedsgrader, hvilket er en typisk manglende faktor i andre kontrol arkitekturer. Den basale hypotese er at en n r relation mellem de visuelle processer og sensor systemet er en n dvendighed for at opn a e ektivitet ved analyse af dynamiske scenarier. Dette har medf rt introduktionen af det s akaldte \Basale System", der varetager kontrollen af sensor systemet p a baggrund af basale visuelle processer. Disse processer er: xation, tracking og attention selektion og skift, hvilket kan udf res udelukkende som data drevet respons. Det basale vision system kan ligeledes servicere foresp rgsler fra h jere niveau visuelle processer, der typisk involvere objekt genkendelse. Disse processer kan kun adressere sensor systemet via det basale system, hvorved en kon ikt om hardware kontrollen undg as. Den designede arkitektur er baseret p a inspiration fra det motoriske system i det menneskelige syn, hvorfra en s akaldt Cyclopisk repr sentation af information er a edt. Denne er n rt knyttet til f nomenet \lige-v gtet" xations kontrol, dvs. jnene samarbejder om hvor og hvad der skal analyseres. Denne repr sentation medf rer at kontrollen af xationspunktet udf res ved styring af de tre vinkler vergence, version og tilt. Ved brug af den Cyclopiske repr sentation separeres position og afstands kontrol, hvilket har v ret benyttet ved fusion af de 2 afstands \cues" accommodering og disparitet til kontrol af de to tilh rende frihedsgrader, focus afstand og vergence vinkel. Dette har gennem eksperimenter vist sig vellykket til at h andtere midlertidig okkludering af det ene kamera. En skalarums baseret attention mekanisme der benytter multiple features er blevet designet og implementeret i arkitekturen. De individuelle processer s avel som det samlede system er blevet analyseret gennem en r kke eksperimenter. Tracking og smooth pursuit er opn aet b ade for stilistiske og mere realistiske objekter. 3D tracking med 10 12Hz har med succes v ret istand til at f lge objekter der bev ger sig mere end 7 =s. En gennemsnitligt n jagtighed bedre end 1cm i en afstand op til 3m er rapporteret. Derudover foretager systemet billedstabilisering, dog ikke for objekter der bev ger hurtigere end 7 =s. Arkitekturen der er designet gennem dette projekt har vist som en e ektiv m ade at organisere de laveste niveauer af et xationsbaseret vision system p a. Dette er hovedsageligt opn aet gennem en separation af de speci kke algoritmer der anvendes og arkitekturen selv. v Acknowledgements While deeply involved in details of this research I received help from or was in other ways supported by a number of people. I wish to express my sincere gratitude to all of them;a few people deserve a special mentioning. I owe the warmest thanks to my supervisor Dr. Henrik Iskov Christensen for his ideas, encouragements and for being patient with me while I was making little progress. I also thank Professor Erik Granum for providing the environment within which this work was carried out, and for the discussions, which have been valuable for my work. Being a member of the Laboratory of Image Analysis at Aalborg University has been a rewarding experience in every respect. For being friends as well as colleagues I thank especially Claus Br ndgaard Madsen, Lars Fr kj r Knudsen, Jan Nielsen and Niels Ole Staub Kirkeby. A substantial portion of the ideas expressed in this thesis were conceived during a six month visit to the University of Rochester, New York. I am deeply indebted to Professor Chris Brown for making this visit a reality. While in Rochester I enjoyed sharing an o ce with Alfonso Gerevini, and had many fun moments playing Softball with him and Ray, Polly, Chris as well as many others. We excelled in performing reasonable even after a fair amount of beer. This work has been sponsored by the Danish Technical Research Council, through the MOBS framework programme, grant no. 5.26.09.07, 5.26.09.07.3, 5.26.09.07.4 and 26.09.07.5. In addition the project has received funding from the Danish Research Academy, the EC through the project EP-7108-VAP, the HCM SMART Network, and an HCM fellowship. vi
منابع مشابه
Modeling and Simulation of Claus Unit Reaction Furnace
Reaction furnace is the most important part of the Claus sulfur recovery unit and its performance has a significant impact on the process efficiency. Too many reactions happen in the furnace and their kinetics and mechanisms are not completely understood; therefore, modeling reaction furnace is difficult and several works have been carried out on in this regard so far. Equilibrium models are co...
متن کاملAn alternative mathematical modeling of the scintillation camera and framework for performance analysis of gamma-ray positioning algorithms
Introduction: Substantive amount of work has been done in modeling and analyzing the scintillation camera system processes including positioning and image formation. The goal of this work is to develop a framework for analyzing performance of nonlinear positioning methods upon construction of a mathematical model of the system. In this study, the photodetector array counts are ...
متن کاملVoltage-Base Control of Camera Stabilizer Using Optimal Adaptive Fuzzy Sliding Mode Control
The camera stabilizer stabilizes the camera’s line of sight by isolating the camera from the model uncertainties, disturbances of operating environment and system movements. This paper presents a voltage-base optimal adaptive fuzzy sliding mode control for camera stabilizer. In this proposed control method, a voltage-base sliding mode controller is applied. But unfortunately, undesirable ...
متن کاملA camera based calculation of 99mTc-MAG3 clearance using conjugate views method [Persian]
Background: Measurement of absolute or differential renal function using radiotracers plays an important role in the clinical management of various renal diseases. Gamma camera quantitative methods in approximation of renal clearance may potentially be as accurate as plasma clearance methods. However some critical factors such as kidney depth and background counts are still troublesome in...
متن کاملComputer Vision Based Head Tracking from Re-configurable 2D Markers for AR
This paper presents a computer vision based head tracking system for augmented reality. A camera is attached to a head mounted display and used to track markers in the user’s field of view. These markers are movable on a table and used as interface with virtual objects. Furthermore, they are used as landmarks to track the user’s viewpoint and viewing direction by a homography based camera pose ...
متن کاملKeyboard before Head Tracking Depresses User Success in Remote Camera Control
In remote mining, operators of complex machinery have more tasks or devices to control than they have hands. For example, operating a rock breaker requires two handed joystick control to position and fire the jackhammer, leaving the camera control to either automatic control or require the operator to switch between controls. We modelled such a teleoperated setting by performing experiments usi...
متن کامل