Tracking with the EM Contour Algorithm

Abstract

A novel active-contour method is presented and applied to pose refinement and tracking. The main innovation is that no "features" are detected at any stage: contours are simply assumed to remove statistical dependencies between pixels on opposite sides of the contour. This assumption, together with a simple model of shape variability of the geometric models, leads to the application of an EM method for maximizing the likelihood of pose parameters. In addition, a dynamical model of the system leads to the application of a Kalman filter. The method is demonstrated by tracking motor vehicles with 3-D models.

Comparison of Edge-Driven Algorithms for Model-Based Motion Estimation

“…Using a single model without wheelarches for all three vehicles yields similar results except for wrong directional estimates of the second vehicle. The results are essentially compatible with those reported in [5].…”

“…The results, as displayed in Fig. 1, show successful tracking of all 3 vehicles if vehicles are initialized interactively, vehicle models are assigned individually, and the same parameters are used as introduced in [5]. Using a single model without wheelarches for all three vehicles yields similar results except for wrong directional estimates of the second vehicle.…”

“…According to the publication [5], the assumptions underlying this algorithm are 1. Greyvalue differences of adjacent pixels have a prior probability distribution…”

“…The one reported in [5,6] is based on a hypothesized probability distribution function for step-discontinuities in the vicinity of model-segments. This approach will be referred to in the sequel as 'Expectation Maximisation Contour Algorithm (EMCA)'.…”

Initialization of Model-Based Vehicle Tracking in Video Sequences of Inner-City Intersections

“…We attempt, however, to at least mention some important publications treating system aspects beyond our current focus. In a notable exception from this scarcity of publications concerned with 3D-model-based tracking in monocular video sequences, Pece and Worrall (2002), see too Pece and Worrall (2006), postulate a probability distribution of Edge Element (EE) locations around model segments which have been projected into the image plane according to the current vehicle state estimate. The state estimate is updated by maximizing the postulated likelihood for observing grayvalue transitions.…”