260 likes | 387 Views
‘’ Pedestrian Tracking Using DCM and Image Correlation ’’. G.Antonini S.Venegas JP.Thiran and M.Bierlaire IM2-2004. Outline. Introduction ( motivations and objectives ) DCM for pedestrian dynamic - Pedestrian behavior modeling (overview) - DCM specification
E N D
‘’ Pedestrian Tracking Using DCM and Image Correlation ’’ G.Antonini S.Venegas JP.Thiran and M.Bierlaire IM2-2004
Outline Introduction ( motivations and objectives ) DCM for pedestrian dynamic - Pedestrian behavior modeling (overview) - DCM specification - DCM calibration - DCM estimation results Pedestrian detection using DCM and image correlation Pedestrian tracking using DCM and image correlation Results Conclusions and future works
Introduction Motivation: new research project conducted with the aim to integrate state-of-the-art tracking algorithms with behavioral models for pedestrian dynamic for video surveillance applications (IM2.SA). Objectives: our goal is to provide a tool for the computation of pedestrian trajectories in real, complex scenarios. These trajectories could then be used to build statistical density maps and land-use maps for scene analysis.
Pedestrian behavior : overview • Previous approaches are mainly physical-based models: people as particles (microscopic models) subjected to forces; people with fluid-like properties (macroscopic models, Navier-Stokes or Boltzmann-like equations). • Our approach: • - microscopic model (time-based behavior of each pedestrian); • - walking is a sequence of choices: where to put the next step? (DCM); • - dynamical and individual-based spatial discretization.
The space model RoI P Static pedestrian area
Dynamic and individual-based We discretize the space discretizing changes in speed module and direction The space model 10 10 10 15 Accelerated 20 Constant speed 25 Decelerated
DCM Choice Set Random variable Alternative’s attributes Socio-economic attributes Behavioral model DCM are disaggregate behavioral models designed to forecast the behavior of individuals in choice situations Choice Set Random variable Alternative’s attributes dm attributes
j 1 11 22 12 23 33 i Behavioral model Choice Set Alternative’s attributes dm attributes Random variable
current direction d destination D Cj O Behavioral model Choice Set Alternative’s attributes dm attributes Random variable
3 k d_kj d_3j 2 d_2j Cj d_1j 1 i Behavioral model Choice Set Alternative’s attributes dm attributes Random variable
Dummy variables capturing the attractiveness of acc / dec. We postulate they vary with the current speed of dm Behavioral model Choice Set Alternative’s attributes dm attributes Random variable
Correlation based on speed Correlation based on direction ACC C CONST NOT C NOT C DEC Behavioral model : the CNL formulation Choice Set Alternative’s attributes dm attributes Random variable
17 6 28 … ,,, 2 10 1 11 12 22 The frequency of choices 23 33 …looking inside collected data… We manually track 36 pedestrians for a total of 1410 position observations
1-1.5345e-00+5.2903e-01 - 2.9005e+00 2-7.9454e-02 +5.2485e-03 - 1.5138e+01 3- 5.3839e-02 +1.9406e-08 - 2.7742e+06 4- 2.4751e+01 + 4.7804e+00 - 5.1775e+00 5- 2.5543e-01 + 5.8682e-02 - 4.3528e+00 6+1.6859e+00 + 1.3083e-01 + 1.2886e+01 8+1.0805e+00+1.3879e-01 +7.7848e+00 +5.8025e-01 11+2.4714e+01 +8.0513e-02 +3.0695e+02 +2.9453e+00 9+2.5210e+00 +6.0733e-01 +4.1509e+00 +2.5044e+00 10+1.4130e+00 +9.0226e-02 +1.5660e+01 +4.5775e+00 7- 8.9047e-01 + 1.3054e-01 - 6.8212e+00 Estimation results: CNL Variable number Variable name Asymptotic std error Coefficients estimates t-test 0 t-test 1 12+1.3271e+00 +1.8426e-01 +7.2024e+00 +1.7754e+00 Rho-square = - 0.484629 Init log_lik. = - 4979.03 Final log_lik. = - 2566.05
Pedestrian simulator Developed by Mats Weber in the context of the CTI project SIMBAD Is initialized with a time-dependent origin-destination matrix. An itinerary is associated with each agent. At each step the utilities and probabilities are calculated (red=high utility, blue=low utility)
Projection of hypothetical moving objects (first frame) Detection of new moving objects in the active zones (each refresh period) camera An hypothetical moving object We assume an average target height of 1.70m Grid on top-view plane Pedestrian detection
Projection of hypothetical moving objects (first frame) camera An hypothetical moving object Original image Background image We assume an average target height of 1.70m Grid on top-view plane Pedestrian detection Foreground image
Next frame Correlation matrix correlation target Pedestrian detection
Pedestrian detection • Pre-filtering: simple thresholding on the visual displacements projected on the top-view plane. An activation value (starting score) is given to each hypthesis. Each bad step consist in a penalty. • Filtering: we use the model’s probabilities to give scores to the trajectories over a period of T frames
Unfiltered trajectories: All trajectories Filtered trajectories: Accepted trajectories Frame 20 Frame 35 Frame 50 Frame 65 Trajectories filtering and detection results
Pedestrian tracking The first approach is to treat tracking as a sequence of detection cycles: deterministic template matching and behavioral filtering.
Pedestrian tracking We use the model as a prior and the normalized image correlation as likelihood (at each step the model is propagated from a MAP estimation on the previous posterior)
To do: • Better representations for the posterior distribution and the likelihood term; • DCM has to be extended to high density scenarios with an explicit model for • fixed and moving obstacles; • We are currently working on a post-clustering of trajectories to integrate at the • end of each detection step. Interesting preliminary results for pedestrians’ • calculation. Conclusions & future works • DCM are flexible and efficient for pedestrian modeling; • The use of behavioral models is usefull both for detection and tracking. Can help • to solve occlusion and illumination condition related problems.