Model-Based Visual Localisation Of Contours And Vehicles
Daniel Ponsa Mussarra · TDX (Tesis Doctorals en Xarxa) · 2007
El treball d'aquesta tesi es centra en l'analisi de sequencies de video, aplicant tecniques basades en models per extreure'n informacio quantitativa. En concret, es realitzen diferents propostes en dues arees d'aplicacio: el seguiment de formes basat en models de contorns, i la deteccio i seguiment de vehicles en imatges proveides per una camera instal·lada en una plataforma mobil. El treball dedicat al seguiment de formes s'enquadra en el paradigma de contorns actius, del qual presentem una revisio de les diferents propostes existents. En primer lloc, mesurem el rendiment obtingut pels algorismes de seguiment mes comuns (filtres basats en Kalman i filtres de particules), i en segon lloc avaluem diferents aspectes de la seva implementacio en un extens treball experimental on es consideren multiples sequencies sintetiques, distorsionades amb diferents graus de soroll. Aixi, mitjancant aquest estudi determinem la millor manera d'implementar a la practica els algorismes de seguiment classics, i identifiquem els seus pros i contres. Seguidament, el treball s'orienta cap a la millora dels algoritmes de seguiment de contorns basats en filtres de particules. Aquest algorismes aconsegueixen bons resultats sempre que el numero de particules utilitzades sigui suficient, pero malauradament la quantitat de particules requerides creix exponencialment amb el numero de parametres a estimar. Per tant, i en el context del seguiment de contorns, presentem tres variants del filtre de particules classic, corresponents a tres noves estrategies per tractar aquest problema. En primer lloc, proposem millorar el seguiment de contorns mirant de propagar mes acuradament les particules emprades per l'algorisme d'una imatge a la seguent. Aixo ho duem a terme utilitzant una aproximacio lineal de la funcio de propagacio optima. La segona estrategia proposada es basa en estimar part dels parametres de manera analitica. Aixi, es preten fer un us mes productiu de les particules emprades, reduint la part dels parametres del model que s'han d'estimar amb elles. El tercer metode proposat te com a objectiu treure profit del fet de que, en aplicacions de seguiment de contorns, sovint els parametres relatius a la transformacio rigida es poden estimar prou acuradament independentment de la deformacio local que el contorn presenti. Aixo s'utilitza per realitzar una millor propagacio de les particules, concentrant-les mes densament en la zona on el contorn seguit es troba. Aquestes tres propostes es validen de manera extensiva en sequencies amb diferents nivells de soroll, amb les que es mesura la millora aconseguida. A continuacio proposem tractar directament l'origen del problema anterior mitjancant la reduccio del nombre de parametres a estimar per tal de seguir una determinada forma d'interes. Per aconseguir aixo, proposem modelar aquesta forma usant multiples models, on cadascun requereix una quantitat de parametres inferior a la requerida per un unic model. Es proposa un nou metode per aprendre aquests models a partir d'un conjunt d'entrenament, aixi com un nou algorisme per emprar-los en el seguiment dels contorns. Els resultats experimentals certifiquen la validesa d'aquesta proposta. Finalment, la tesi es centra en el desenvolupament d'un sistema de deteccio i seguiment de vehicles. Les propostes realitzades comprenen: un modul de deteccio de vehicles, un modul dedicat a determinar la posicio i velocitat 3D dels vehicles detectats, i un modul de seguiment per actualitzar la localitzacio dels vehicles a la carretera de manera precisa i eficient. Es realitzen diverses aportacions originals en aquests tres temes, i se n'avalua el rendiment. This thesis focuses the analysis of video sequences, applying model-based techniques for extracting quantitative information. In particular, we make several proposals in two application areas: shape tracking based on contour models, and detection and tracking of vehicles in images acquired by a camera installed on a mobile platform. The work devoted to shape tracking follows the paradigm of active contours, from which we present a review of the existent approaches. First, we measure the performance of the most common algorithms (Kalman based filters and particle filters), and then we evaluate its implementation aspects trough an extensive experimental study, where several synthetic sequences are considered, distorted with different degrees of noise. Thus, we determine the best way to implement in practice these classical tracking algorithms, and we identify its benefits and drawbacks. Next, the work is oriented towards the improvement of contour tracking algorithms based on particle filters. These algorithms reach good results provided that the number of particles is high enough, but unfortunately the required number of particles grows exponentially with the number of parameters to be estimated. Therefore, and in the context of contour tracking, we present three variants of the classical particle filter, corresponding to three new strategies to deal with this problem. First, we propose to improve the contour tracking by propagating more accurately the particles from one image to the next one. This is done by using a linear approximation of the optimal propagation function. The second proposed strategy is based in estimating part of the parameters analytically. Thus, we aim to do a more productive use of the particles, reducing the amount of model parameters that must be estimated through them. The third proposed method aims to exploit the fact that, in contour tracking applications, the parameters related to the rigid transform can be estimated accurately enough independently from the local deformation presented by the contour. This is used to perform a better propagation of the particles, concentrating them more densely in the zone where the tracked contour is located. These three proposals are validated extensively in sequences with different noise levels, on which the reached improvement is evaluated. After this study, we propose to deal directly with the origin of the previous problem by reducing the number of parameters to be estimated in order to follow a given shape of interest. To reach that, we propose to model the shape using multiple models, where each one requires a lower quantity of parameters than when using a unique model. We propose a new method to learn these models from a training set, and a new algorithm to use the obtained models for tracking the contours. The experimental results certify the validity of this proposal. Finally, the thesis focuses on the development of a system for the detection and tracking of vehicles. The proposals include: a vehicle detection module, a module devoted to the determination of the three-dimensional position and velocity of the detected vehicles, and a tracking module for updating the location of vehicles on the road in a precise and efficient manner. Several original contributions are done in these three subjects, and their performance is evaluated empirically.