Norbert Brändle

Pedestrian Detection and Tracking for Counting Applications in Crowded Situations

This paper describes a vision based pedestrian detection and tracking system which is able to count people in very crowded situations like escalator entrances in underground stations. The proposed system uses motion to compute regions of interest and prediction of movements, extracts shape information from the video frames to detect individuals, and applies texture features to recognize people. A search strategy creates trajectories and new pedestrian hypotheses and then filters and combines those into accurate counting events. We show that counting accuracies up to 98 % can be achieved.

Estimating Pedestrian Movement Characteristics for Crowd Control at Public Transport Facilities

Capacities of doors, staircases and other bottle-necks are a key aspect in the design of infrastructures for public transport. Especially major events like soccer games and concerts may lead to large crowds which need to be accommodated, while at the same time potential safety hazards like overcrowding must be avoided. The bottleneck capacities limit the capacities of the whole system and control the potential of high crowd densities on critical elements such as the platforms. We present an approach to estimate the maximum and effective capacity of key bottleneck elements based on controlled experiments and real world data sets of pedestrian movements for a subway station next to the main soccer stadium in Vienna. The focus is the fundamental diagram revealing both the maximal capacity as well as the effective capacity in terms of pedestrian flow rates. We present two controlled experiments and results based on real world data obtained during the European Soccer Championship (UEFA EURO 2008trade).

Track-Based Finding of Stopping Pedestrians - A Practical Approach for Analyzing a Public Infrastructure

We present a case study for obtaining and analyzing long-term pedestrian track data within a large hall of an Austrian railway station, where no CCTV surveillance system was pre-installed. Hence one focus of this paper concerns practical aspects for selecting and strategically placing a high-quality multi-camera system for recording long-term video data for off-line analysis. The pedestrian tracks are obtained by applying a state-of-the art vision-based multiple human tracking software program. The focus of the track analysis is the detection of places where people stop frequently or walk slowly, respectively. We present the applied methodology and discuss the main results

Evaluation of clustering methods for finding dominant optical flow fields in crowded scenes

Video footage of real crowded scenes still poses severe challenges for automated surveillance. This paper evaluates clustering methods for finding independent dominant motion fields for an observation period based on a recently published real-time optical flow algorithm. We focus on self-tuning spectral clustering and Isomap combined with k-means. Several combinations of feature vector normalizations and distance measures (Euclidean, Mahanalobis and a general additive distance) are evaluated for four image sequences including three publicly available crowd datasets. Evaluation is based on mean accuracy obtained by comparison with a manually defined ground truth clustering. For every dataset at least one approach correctly classified more than 95% of the flow vectors without extra tuning of parameters, providing a basis for an automatic analysis after a view-dependent setup.

Robust road link speed estimates for sparse or missing probe vehicle data

Probe vehicles equipped with GPS can be used to permanently collect traffic speed information for an entire road network, and the statistical mean value of link speeds collected over time is often used as an estimator for mid-term predictions. For road links with sparse probe vehicle data, the estimated mean may be too inaccurate due to the low sample size, and speeds for road links with missing probe vehicle data must be imputed from other data. This paper proposes to apply a Gaussian-mixture based technique to increase the robustness of speed estimates. Typical shapes of the diurnal speed curve are learnt from historical data of all links in the road network. The model is able to provide robust estimates of mean speed curves based on only a few available observations and drastically reduces the amount of data needed to store them by 95%. Experimental results on a comprehensive set of 857527 day speed curves show that the predictions are superior to traditional approaches based on aggregated or disaggregated historical mean values.

Next-generation 3D visualization for visual surveillance

Existing visual surveillance systems typically require that human operators observe video streams from different cameras, which becomes infeasible if the number of observed cameras is ever increasing. In this paper, we present a new surveillance system that combines automatic video analysis (i.e., single person tracking and crowd analysis) and interactive visualization. Our novel visualization takes advantage of a high resolution display and given 3D information to focus the operators attention to interesting/ critical areas of the observed area. This is realized by embedding the results of automatic scene analysis techniques into the visualization. By providing different visualization modes, the user can easily switch between the different modes and can select the mode which provides most information. The system is demonstrated for a real setup on a university campus.

Learning Major Pedestrian Flows in Crowded Scenes

We present a crowd analysis approach computing a representation of the major pedestrian flows in complex scenes. It treats crowds as a set of moving particles and builds a spatio-temporal model of motion events. A Growing Neural Gas algorithm encodes optical flow particle trajectories as sequences of local motion events and learns a topology which is the base for trajectory distance computations. Trajectory prototypes are aligned with a two-open-ends version of Dynamic Time Warping to cope with fragmented trajectores. The trajectories are grouped into an automatically determined number of clusters with self-tuning spectral clustering. The clusters are compactly represented with the help of Principal Component Analysis, providing a technique for unusual motion detection based on residuals. We demonstrate results for a publicly available crowded video and a scene with volunteers moving according to defined origin-destination flows.

Design of Decision Rules for Crowd Controlling Using Macroscopic Pedestrian Flow Simulation

Crowd control mechanisms such as temporary access restrictions allow affecting pedestrian flows in public transport facilities in order to avoid overcrowding. Such access restrictions can be based on decision rules depending on measured pedestrian density. In order to design these decision rules, simulations of pedestrian flows are a valuable tool. In this paper we describe the design of decision rules for a real case study constituted by a subway station next to the main soccer stadium in Vienna. The simulations use a macroscopic model which (1) includes dynamic elements (like e.g. arriving and departing trains) and (2) integrates the implementation of decision rules based on real time measurements of real people flows. The model also takes into account measurement errors. We discuss the simulation results for the case study with the resulting decision rules in place.

Finding Highly Frequented Paths in Video Sequences

We propose a novel algorithm to find highly frequented paths of motion trajectories obtained from video sequences. This is achieved by representing the motion trajectories in the scene as sequences of prototypes obtained by a combined vector quantization and growing neural gas algorithm. In contrast to existing methods, the proposed algorithm can be applied to data sets containing motion trajectories of varying length. The algorithm does not assume an a priori fixed number of prototypes. We demonstrate results on surveillance video sequences of cars driving on a highway and pedestrians walking in a major railway station

Assessing traffic performance using position density of sparse FCD

We present an approach for evaluating traffic performance along corridors and its variation based on floating car data (FCD). In contrast to existing work, our approach can cope with long and irregular FCD reporting intervals. Resampling of sparse FCD in time and interpolation increases spatial resolution of FCD positions along the corridors. FCD position density is computed with a uniform kernel, which leads to traffic performance expressed as average travel time per meter and average speed. Experimental results based on real-world FCD for a freeway section and arterial roads in Vienna illustrate the plausibility of the approach, and an example illustrating our approach before and after a traffic influencing measure shows its advantage over using dedicated probe vehicle runs, temporary sensor installations or human observers. A sensitivity analysis provides guidelines for the important parameters.