Henner Kollnig

3D Pose Estimation by Directly Matching Polyhedral Models to Gray Value Gradients

This contribution addresses the problem of pose estimation and tracking of vehicles in image sequences from traffic scenes recorded by a stationary camera. In a new algorithm, the vehicle pose is estimated by directly matching polyhedral vehicle models to image gradients without an edge segment extraction process. The new approach is significantly more robust than approaches that rely on feature extraction since the new approach exploits more information from the image data. We successfully tracked vehicles that were partially occluded by textured objects, e.g., foliage, where a previous approach based on edge segment extraction failed. Moreover, the new pose estimation approach is also used to determine the orientation and position of the road relative to the camera by matching an intersection model directly to image gradients. Results from various experiments with real world traffic scenes are presented.

3D pose estimation by fitting image gradients directly to polyhedral models

Addresses the problem of pose estimation and tracking of vehicles in image sequences from traffic scenes recorded by a stationary camera. In a new algorithm, the vehicle pose is estimated by directly fitting image gradients to polyhedral vehicle models without an edge segment extraction process. The new approach is significantly more robust than approaches that rely on feature extraction because the new approach exploits more information from the image data. We can track vehicles that are partially occluded by textured objects, e.g. foliage, where classical approaches based on edge segment extraction fail. Results from various experiments with real-world traffic scenes are presented.<<ETX>>

Association of motion verbs with vehicle movements extracted from dense optical flow fields

This contribution addresses the problem of detection and tracking of moving vehicles in image sequences from traffic scenes recorded by a stationary camera. By replacing the low level vision system component for the estimation of displacement vectors by an optical flow estimation module we are able to detect all moving vehicles in our test image sequence. By replacing the edge detector and by doubling the sampling rate we improve the model-based object tracking system significantly compared to an earlier system. The trajectories of vehicles are characterized by motion verbs and verb phrases. Results from various experiments with real world traffic scenes are presented.

Tracking of Occluded Vehicles in Traffic Scenes

Vehicles on downtown roads can be occluded by other vehicles or by stationary scene components such as traffic lights or road signs. After having recorded such a scene by a video camera, we noticed that the occlusion may disturb the detection and tracking of vehicles by previous versions of our computer vision approach. In this contribution we demonstrate how our image sequence analysis system can be improved by an explicit model-based recognition of 3D occlusion situations. Results obtained from real world image sequences recording gas station traffic as well as inner-city intersection traffic are presented.

Motion boundary detection in image sequences by local stochastic tests

While estimating both components of optical flow based on the postulated validity of the Optical Flow Constraint Equation (OFCE), it has been tacitly assumed so far that the partial derivatives of the gray value distribution — which are required for this approach at the pixel positions involved — are independent from each other. [Nagel 94] has shown in a theoretical investigation how dropping this assumption affects the estimation procedure. The advantage of such a more rigorous approach consists in the possibility to replace heuristic tests for the local detection of discontinuities in optical flow fields by well known stochastic tests. First results from various experiments with this new approach are presented and discussed.

Matching Object Models to Segments from an Optical Flow Field

The temporal changes of gray value structures recorded in an image sequence contain significantly more information about the recorded scene than the gray value structures of a single image. By incorporating optical flow estimates into the measurement function, our 3D pose estimation process exploits interframe information from an image sequence in addition to intraframe aspects used in previously investigated approaches. This increases the robustness of our vehicle tracking system and facilitates the correct tracking of vehicles even if their images are located in low contrast image areas. Moreover, partially occluded vehicles can be tracked without modeling the occlusion explicitly. The influence of interframe and intraframe image sequence data on pose estimation and vehicle tracking is discussed systematically based on various experiments with real outdoor scenes.

Influence of an Explicitly Modelled 3D Scene on the Tracking of Partially Occluded Vehicles

Model-based tracking of vehicles in real world image sequences of traffic may fail due to different reasons. A careful analysis of failed tracking experiments brought to light that one of these phenomena consists in an incorrect match of parts of the vehicle model to image features belonging to other scene components. This effect appears in particular if the vehicle is occluded by either stationary scene components or by another moving vehicle. Although we got some encouraging results when we modeled the occluding scene components explicitly, we still encountered cases in which we did not succeed in tracking partially occluded vehicles properly. In this contribution we show some successful tracking results obtained from real world image sequences and discuss the cases of failures. Additionally, we present a framework for the conceptual characterization of occurring occlusions which became amenable to experimental analysis in the course of our investigations.

Zuordnung natürlichsprachlicher Begriffe zu Geschehen an einer Tankstelle

Das in diesem Beitrag vorgestellte Bildauswertungssystem berechnet begriffliche Beschreibungen fur automatisch aus Videobildfolgen ermittelte Trajektoriendaten. Dabei sind alle notwendigen Schritte von den Sensordaten uber geometrische bis hin zu begrifflichen Beschreibungen in einem einheitUchen System implementiert. Ausgehend von den theoretischen Uberlegungen in [Nagel 91] werden zulassige Aktionssequenzen beim Betanken eines Fahrzeuges weiterentwickelt und an — aus den Bilddaten ermittelte — Geschehensbeschreibungen gekoppelt. Damit stehen reichhaltigere begriffliche Beschreibungen zur Verfugung als bisher.

Evaluation of Image Sequences from Outdoor Scenes: Selected Problems and Solutions

Lane structures are estimated, based on image sequences recording complex inner-city intersections, by a model-based approach. First, by estimating vanishing point coordinates from straight line segments extracted from the image sequence, an initial guess regarding the orientation of the camera with respect to the road plane as well as regarding the focal length is obtained. Subsequently, various steps of both model and process refinement result in a significantly improved lane structure estimation. Results from experiments with real world innercity multi-lane intersection scenes are presented.

Detektion und Verfolgung von Fahrzeugen in Straßenverkehrsszenen: Systematische Bewertung und Steigerung der Systemleistung

Die Leistung eines Bildfolgenauswertungssystems wird durch Einsatz von mehr Modellwissen signifikant verbessert. Die Segmentierung eines optischen Flusfeldes unter Verwendung von Wissen uber den 3D-Szenenbereich liefert genauere Hinweise auf Abbilder bewegter Fahrzeuge als Verfahren, die rein im 2D-Bildbereich arbeiten. Dadurch last sich die Verfolgung von Fahrzeugen besser initialisieren. Die automatische Detektion von Schwierigkeiten bei der Verfolgung von Fahrzeugen sowie eine gegebenenfalls automatisch veranlaste Neuinitialisierung des Verfolgungsprozesses fuhrt zu einer weiteren Leistungssteigerung des Gesamtsystems. Die Qualitat der Ergebnisse rechtfertigt den Aufwand fur eine quantitative Bewertung der Systemleistung auf umfangreicheren Datensatzen.