Zoran Kalafatić

Infrared-visual image registration based on corners and hausdorff distance

The paper presents an approach to multimodal image registration. The method is developed for aligning infrared (IR) and visual (RGB) images of facades. It is based on mapping clouds of points extracted by a corner detector applied to both images. The experiments show that corners are suitable features for our application. In the alignment process a number of transformation hypotheses is generated and evaluated. The evaluation is performed by measuring similarity between the RGB corners and the transformed corners from IR image. Directed partial Hausdorff distance is used as a robust similarity measure. The implemented system has been tested on various IR-RGB pairs of images of buildings. The results show that the method can be used for image registration, but also expose some typical problems.

Exploiting temporal and spatial constraints in traffic sign detection from a moving vehicle

This paper addresses detection, tracking and recognition of traffic signs in video. Previous research has shown that very good detection recalls can be obtained by state-of-the-art detection algorithms. Unfortunately, satisfactory precision and localization accuracy are more difficultly achieved. We follow the intuitive notion that it should be easier to accurately detect an object from an image sequence than from a single image. We propose a novel two-stage technique which achieves improved detection results by applying temporal and spatial constraints to the occurrences of traffic signs in video. The first stage produces well-aligned temporally consistent detection tracks by managing many competing track hypotheses at once. The second stage improves the precision by filtering the detection tracks by a learned discriminative model. The two stages have been evaluated in extensive experiments performed on videos acquired from a moving vehicle. The obtained experimental results clearly confirm the advantages of the proposed technique.

Restoration of images blurred by circular motion

The problem of restoring images blurred by circular motion is considered. In order to simplify the process of deblurring, the original image in the (x,y) plane is transformed into polar plane (p,/spl phi/) . This simplifies the blur model, resulting in one-dimensional integration.

Generative modeling of spatio-temporal traffic sign trajectories

We consider the task of automatic detection and recognition of traffic signs in video. We show that successful off-the-shelf detection (Viola-Jones) and classification (SVM) systems yield unsatisfactory results. Our main concern are high false positive detection rates which occur due to sparseness of the traffic signs in videos. We address the problem by enforcing spatio-temporal consistency of the detections corresponding to a distinct sign in video. We also propose a generative model of the traffic sign motion in the image plane, which is obtained by clustering the trajectories filtered by an appropriate procedure. The contextual information recovered by the proposed model will be employed in our future research on recognizing traffic signs in video.

A system for tracking laboratory animals based on optical flow and active contours

We present a working system for real-time tracking of multiple laboratory animals. As it is usually possible to ensure good contrast between the animals and the background, the tracking of a single animal or several physically separated animals can be obtained by relatively simple algorithms. The main problem arises when we try to track several almost identical, uniformly coloured animals during their contacts. To deal with this problem, we utilize dynamic information extracted by estimating sparse optical flow along the object contours. Optical flow vectors are used for updating the positions of the tracked contours in a sequence of image frames. The local properties of optical flow enable the system to track the objects during their contact, although some parts of the object contours become hidden. The missing dynamic information is reconstructed by using a model of constant optical flow along an object contour. The reconstructed contours are then adjusted to real object boundaries in the current frame by using an active contour model. The robustness of the tracking algorithm is improved by adding a supervision module, which detects tracking failures and reinitialises the contours that lose their targets. The system has been tested on real sequences with laboratory animals during pharmacological experiments and has been shown to be robust and efficient. Future extensions will include expert knowledge of biomedical and pharmacological experts. The major goal is to build a system that will provide an objective and standardised tool for evaluation of animal behaviour during experiments.

A computer vision assisted geoinformation inventory for traffic infrastructure

Geoinformation inventories are often employed as a tool for providing a comprehensive view onto the required state of traffic control infrastructure. They are especially important in road safety inspection where, in combination with georeferenced video, they enable repeatable off-line and off-site assessments as an attractive aternative to classic onsite inspection. Nevertheless, manual assessments are tedious and time-consuming even when performed off-line, and this seriously impairs the potential of the geoinformation inventory concept. This paper therefore researches a hypothesis that suitable georeferenced video processing techniques would allow reliable automation of the following operations: i) creation of the traffic inventory from the given video, and ii) assessing the video against the state in the inventory. Prominent computer vision approaches have been rigorously and systematically evaluated and the obtained results are presented. The results seem to support the hypothesis, although further work is required for a more definite answer.

Model-based tracking of laboratory animals

We present a system for tracking laboratory animals during pharmacological experiments. As it is usually possible to ensure good contrast between the animals and the background, tracking of a single animal or several physically separated animals can be achieved by relatively simple algorithms. The main problem arises when we try to track several almost identical, uniformly coloured animals during their contacts. To deal with this problem we represent objects by parametrically deformable contour models. The model has been built by observing videos containing a single animal (a laboratory mouse). To reflect symmetry, the model is axial and contains the offsets of the contour segments from the axis of minimal inertia. The deformation is modeled as stretching and bending. The tracking is done in two steps. For the tracking of objects from frame to frame we use the rigidity assumption, i.e. in the first step the contour models which represent objects in the previous frame are translated into new positions. In the second step the object position, rotation and scale, as well as the deformation parameters, are fine-tuned to match the object boundaries. The interframe translation is estimated by minimizing the sum of squared differences (SDD) over the search window for all tracked contour points. The model fitting is based on maximizing the contour energy in terms of the underlying smoothed gradient image. The robustness of the tracking algorithm is improved by adding a supervision module, which detects tracking failures and reinitializes the contours that lose their targets. The system has been tested on real sequences with laboratory animals during pharmacological experiments and has been shown to be robust and efficient. Future extensions will include expert knowledge of biomedical and pharmacological experts. The major goal is to build a system that will provide a tool for objective evaluation of animal behaviour during experiments.

Wireless sensor node modelling for energy efficiency analysis in data-intensive periodic monitoring

Data-intensive wireless sensor applications, such as remote visual inspection using high-resolution video sensors, require a special design approach in order to save energy and prolong lifetime of a battery-powered wireless sensor node. This study is motivated by searching for the most efficient communication protocol for high-resolution image transmission in environmental monitoring sensor networks, where data should be transmitted periodically, but relatively rarely (usually once or twice per day). Some previous publications propose ZigBee or Wi-Fi as suitable candidates for data-intensive wireless transmission, but the literature lacks a systematic study that would provide a guidance for designing such systems. We construct a measurement-based model of a wireless sensor node with emphasis on the communication unit. We measured the energy consumption of commercially available wireless ZigBee and Wi-Fi modules, as well as the influence of the interface bandwidth limitation that reduces their energy efficiency. The model includes real-world communication channel properties that at high bit-rates reduce the communication range and increase the energy consumption due to a higher susceptibility to noise.Our results show that in scenarios when the node sends up to 64kB of data per session once per day, the estimated lifetime of a ZigBee node is up to 10% longer than of a Wi-Fi node. However, when the amount of data per session increases, the Wi-Fi wins due to its higher energy efficiency during data transfer. When the data amount reaches 10MB, the lifetime of a Wi-Fi node using UDP protocol is 5 times longer than that of a ZigBee node. On the other hand, the Wi-Fi node lifetime decreases with increasing number of sessions per day, because the connection establishment with the access point is very energy consuming. As a result, when 5 sessions per day are required the ZigBee node can offer 40% longer lifetime than the Wi-Fi node when 10kB of data is transmitted per session.

Deep metric learning for person Re-identification and De-identification

Large amounts of visual data are gathered from various surveillance systems across different places and times, and have to be processed in order to infer the current state of the world. One of the common problems in surveillance scenarios is person re-identification, the task of associating a person across different cameras. On the other hand, these scenarios raise privacy concerns, which lead to the need for person de-identification, i.e. concealing person identity. This task is related to the re-identification in two aspects: (i) multiple appearances of the same person could be de-identified in similar manner; and (ii) if we discover the features useful for re-identification, we could try to hide the identity by modifying those features. Re-identification can be addressed as a classification problem. The state-of-the-art classification methods are based on deep learning. In this paper we explore the applicability of the recently proposed Triplet network architecture to the person re-identification problem, by applying it on VIPeR dataset. We show that the network is able to learn useful feature-space embeddings, and analyze its benefits and limitations.

Tracking objects using particle filters

This paper describes an implementation of particle filter tracker based on condensation algorithm. The filter processes measurements as they become available in a standard predict-update loop. The prediction phase uses the available dynamic model to predict the probability density function in the next time step, by applying both the deterministic and stochastic component of the model to all samples. In the update phase the new measurement is used to update the probability density function by updating the weight of each sample. The goal of this work was to investigate the possibilities of object tracking without learning a dynamic motion model. Changes to the basic algorithm have been implemented that can help to improve the tracking performance by using more than one motion model and more than one predict-update iteration per measurement.