City Research Online

Abstract

This thesis is concerned with tracking man made objects moving in natural open world scenes and based on the tracking data, construct a structural representation of that scene, frame by frame. The system developed uses a static camera and a statistical frame differencing technique for detecting motion in an image that has a relatively static background. Objects with a measured temporal consistency are tracked across successive image frames. Based on the tracking data, regions in the scene are associated with particular types of dynamic event. For example regions containing movement (could be roads) and regions where objects seem to disappear or partially disappear (could be hedges).

Because of the sensitivity of the motion estimator to changes in scene illumination and environmental conditions, a tile-based method is used to detect scene motion based on the estimations of statistical variations within the tiles. An updating process is used to ensure that a reliable estimate of the background reference image is maintained by the system. Motion cues are matched against tracked objects from a previous frame using an estimate of the temporal continuity of an object. A spatial-temporal reasoning process is used to infer the structure in the image. This inference mechanism is implemented using a semantic network.

The system has been tested on several open world sequences and in each case has demonstrated that it can identify and track vehicles moving in the scene. Based on the motion of these vehicles regions in the image were identified and scene maps constructed for each scene. The map identified regions where vehicles can be expected to be observed moving and regions where they could become occluded.

A CD-ROM is included with this thesis that contains the results obtained by the system for the two image sequences used in chapter seven. These results incorporate some of the enhancements outlined in chapter 8, section 8.3. A windows movie player is included on the CD-ROM and appendix d provides information on the contents of the CD-ROM together with installation and operating instructions.

Publication Type:	Thesis (Doctoral)
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science T Technology > TA Engineering (General). Civil engineering (General)
Departments:	School of Science & Technology > Computer Science School of Science & Technology > School of Science & Technology Doctoral Theses Doctoral Theses

Export

Downloads