Particle filter | EPFL Graph Search

Particle filters, or sequential Monte Carlo methods, are a set of Monte Carlo algorithms used to find approximate solutions for filtering problems for nonlinear state-space systems, such as signal processing and Bayesian statistical inference. The filtering problem consists of estimating the internal states in dynamical systems when partial observations are made and random perturbations are present in the sensors as well as in the dynamical system. The objective is to compute the posterior distributions of the states of a Markov process, given the noisy and partial observations. The term "particle filters" was first coined in 1996 by Pierre Del Moral about mean-field interacting particle methods used in fluid mechanics since the beginning of the 1960s. The term "Sequential Monte Carlo" was coined by Jun S. Liu and Rong Chen in 1998. Particle filtering uses a set of particles (also called samples) to represent the posterior distribution of a stochastic process given the noisy and/or pa

Camera Tracking using natural feature point matching for Augmented Reality

Touradj Ebrahimi, Cecilia Torralbo Gimeno

For augmented reality applications, accurate estimation of the camera pose is required. An existing video-based marker tracking system developed at the ITS presents some limitations when the marker is completely or partially occluded. The objective of this project is to improve the tracking of the camera on those cases. For this purpose, the tracking of the marker is extended to that of natural feature points around the marker's environment. The extension to natural features works as follows. Feature point candidates are being searched in the image. Depth estimation is implemented in a recursive method triangulating points between the first frame and subsequent frames. A particle filter is used to take into account the new measures received with every frame. Particle's weights are sensible to the state of the camera tracker and the proximity of the triangulated point to the expected depth. Eventually the distribution of the depth particle filter for a given feature converges at one point. This position is added to a map of feature points used to update the camera pose. Results show that tracking was improved. The camera pose was tracked even though the marker was not visible. However a small drift in its pose is detected due to the imprecision of feature point positions. This drift is not excessive in the sense that the tracker is able to recover the right camera pose. Future work will focus on improving the accuracy of the depth estimates.

2007

Audio-based Positioning and Target Localization for Swarms of Micro Aerial Vehicles

Meysam Basiri

Swarms of autonomous MAVs show great potential in many diverse applications. In a search and rescue mission, a group of MAVs could quickly reach disaster areas by flying over obstacles, cluttered and inaccessible terrains, and work in parallel to detect and locate people that are in need of help. However, multiple challenges still remain in the design of truly autonomous MAV teams. Due to the strict constraints imposed on the MAVs in terms of size, weight, 3D coverage, processing power, power consumption and price, there are not many technological possibilities that could provide individuals with information abut the position of themselves and their team mates without the aid of external systems. Inspired by the sense of hearing among many animal groups which use sound for localization purposes, we propose an on-board audio-based system for allowing individuals in an MAV swarm to use sound waves for obtaining the relative positioning, and furthermore, the self-localization information. We show that not only such a system fully satisfies the constraints of MAVs and is capable of obtaining these information, but also provides additional important opportunities, such as the detection and localization of crucial acoustic targets in the environment. Operating during night time, through foliage and in adverse weather conditions such as fog, dust and smoke, and detection and collision avoidance with non-cooperative noise-emitting aerial platforms are some of the potential advantages of audio-based swarming MAVs. In this thesis, we firstly describe an on-board sound-source localization system and novel methods for the innovative idea of localizing emergency sound sources on the ground from airborne MAVs. The ability of MAVs to locate source of distress sound signals, such as the sound of an emergency whistle blown by a person in need of help or the sound of a personal alarm, is significantly important and would allow fast localization of victims and rescuers during night time and in fog, dust, smoke and dense forests. We propose three different methods for overcoming the ambiguities related with localizing emergency sound sources or, in general, narrowband sound sources. Furthermore, we present the on-board audio-based relative positioning system for obtaining information about the position of other MAVs in their vicinity. We initially describe a passive method that exploits only the engine sound of other robots, in the absence of the self-engine noise, to measure their relative directions. We then extend this method to overcome some of its limitations, where individuals generate a chirping sound similar to the sound of birds to assist others in obtaining their positions. We then describe an estimator based on particle filters that fuses the relative bearing measurements with information about the motion of the robots, provided by their onboard sensors, to also obtain an estimate about the relative range of the robots. Finally, we present a cooperative method to address the self-localization problem for a team of MAVs, while accommodating the motion constraints of flying robots, where individuals obtain their three dimensional positions through perceiving a sound-emitting beacon MAV. All methods are based on coherence testing among signals of a small on-board microphone array, to measure the probable direction of incoming sound waves, and estimators for robust estimation of the desired information throughout time.

EPFL2015

Joint head tracking and pose estimation for visual focus of attention recognition

Silèye Oumar Ba

During the last two decades, computer science what are the ability to give to provide to machines in order to give them the ability to understand human behavior. One of them which is an important key to understand human behaviors, is the visual focus of attention (VFOA) of a person, which can be inferred from the gaze direction. The VFOA of a person gives insight about information such as who or what is the interest of a person, who is the target of the person's speech, who the person is listening to. Our interest in this thesis is to study people's VFOA using computer vision techniques. To estimate the VFOA of attention from a computer vision point of view, it is required to track the person's gaze. Because, tracking the eye gaze is impossible on low or mid resolution images, head orientation can be used as a surrogate for the gaze direction. Thus, in this thesis, we investigate in a first step the tracking of people's head orientation, in a second step the recognition of their VFOA from their head orientation. For the head tracking, we consider probabilistic methods based on sequential Monte Carlo (SMC) techniques. The head pose space is discretized into a finite set of poses, and Multi-dimensional Gaussian appearance models are learned for each discrete pose. The discrete head models are embedded into a mixed state particle filter (MSPF) framework to jointly estimate the head location and pose. The evaluation shows that this approach works better than the traditional paradigm in which the head is first tracked then the head pose is estimated. An important contribution of this thesis is the head pose tracking evaluation. As people usually evaluate their head pose tracking methods either qualitatively or with private data, we built a head pose video database using a magnetic field 3D location and orientation tracker. The database was used to evaluate our tracking methods, and was made publicly available to allow other researchers to evaluate and compare their algorithms. Once the head pose is available, the recognition of the VFOA can be done. Two environments are considered to study the VFOA: a meeting room environment and an outdoor environment. In the meeting room environment, people are static. People's VFOAs were studied depending on their locations in the meeting room. The set of VFOAs for a person is discretized into a finite set of targets: the other people attending the meeting, the table, the slide screen, and another VFOA target called un-focused denoting that the person is focusing none of the previous defined VFOAs. The head poses are used as observations and potential VFOA targets as hidden states in a Gaussian mixture model (GMM) or a hidden Markov model (HMM) framework. The parameters of the emission probability distributions were learned by two ways. A first way using head pose training data, and a second way exploiting the geometry of the room and the head and eye-in-head rotations. Maximum a posteriori adaptation (MAP) of the VFOA models was to the input test data to take into account people personal ways of gazing at VFOA targets. In the outdoor environment, people are moving and there is a single VFOA target. The problem in this study is to track multiple people passing and estimate whether or not they were focusing the advertisement. The VFOA is modeled as a GMM having as observations people's head location and pose.