Recursive Bayesian estimators for people counting using video sensors
- Publication Type:
- Issue Date:
NO FULL TEXT AVAILABLE. Access is restricted indefinitely. ----- Real time pedestrian flow information and the count of people in determined areas are essential for a multitude of facility management and monitoring functions, including the optimal utilization of floor space. The range of applications is wide and the focus here is on the development of statistical methods for people volume estimation in buildings. To count people, visual information from surveillance cameras is used. This is the first of a series of reports, where the problem is divided in estimation classes of increasing difficulty. There is no widespread use of intelligent systems for surveillance. In addition, most of the existing systems are limited in scope and provide answers only for areas covered by cameras. We propose to develop a generic approach for estimating crowds in large locations, based on partial information. We propose to do so using probabilistic modeling. We first model the number of people in camera covered environments. We then gradually enlarge the scope of the problem to provide an answer for the estimation of the pedestrian flow in locations such as large buildings. In our approach, we propose to develop probabilistic models for the count of people in the views of cameras at time t, N(t), and estimate the number of people in areas not covered by cameras. In the initial stage, probability models are derived for the basic counting scenario, using data gathered from a single camera. The methodology is extended to more complex problems, making use of several cameras covering the same field of view. The consideration of all possible classes of estimation problems leads to solutions that culminate in the assessment of the total number of people in a building. We present a systematic approach to extracting vital statistics like arrival/departure rates and stay times across the facility from visual sensors. We use an object tracker to extract people related data from the video sensors. The number of people in the covered area is determined. The inter-arrival times have Exponential Distribution with mean(λ), this results in the arrival stochastic process being a Poisson process within a homogeneous period. The arrival process throughout the day is the non-homogeneous Poisson process (NHPP). We use a Bayesian approach with a conjugate prior distribution Gamma for λ parameter that leads to a posterior Gamma distribution. Statistics for the stay times are deducted from the stochastic process of the number of incoming people. A Gamma probability model is assumed for the stay times with unknown parameters alpha and beta. The Bayesian methodology is applied to model these parameters probabilistically, using the statistical data.
Please use this identifier to cite or link to this item: