Study Of Moving Objects Detection Based On Spatial-temporal Gradient In Image Sequence

This dissertation was to research and develop the motion target detection technologies based on image sequences. The related technologies mainly included the motion estimation, target detection, false signal suppression, and image filtering, and so on. These key technologies were studied in depth. For different applications, the image sequences were quite different. Therefor, it was difficult to find out a common technology for all. This dissertation developed them according to the different characteristics of target, background and noise. The discussed target was big or small; the background, fixed or moving; and the noise, simple or complex.To the big target detection, the strategy was to analyse the motion field (MF) of the image sequence, estimate and compensate the background motion, and then detect the target. In the image, the complex background motion was represented with a parameter model as the basic motion combination of translation, rotation, and zoom. This dissertation researched the motion estimation (ME) technology based on the correlative block matching (CBM) and/or optical flow equation (OFE). In the CBM-based ME technology, the gradient correlative block (GCB) was proposed. In the OFE-based ME technology, the application conditions were analyzed systematically, and then, to meet the constraints, the improved OFE algorithm was proposed. The improved OFE algorithm was used for the monitoring application of city mobiles, whereafter, a velocity measuring system based on single camera was proposed.To the small target detection, this dissertation adopted the different strategy to image sequences captured by fixed or moving camera. The clutter estimation, suppression technology and the spatial-temporal detection technology based on image gray were used for the fixed background. The ME technology and temporal integrated detection technology based on motion field, were used for the moving background.To the fixed background, this dissertation proposed two technologies of clutter estimation based on image gradient, one based on the minimum cumulating-squared error (MCSE) of gradient, the other based on the maximum correlation of gradients. PC experiments indicated that these technologies of the small targets detection improved the contrast between the target and background, and simplified the gray distribution of the image remainder. Then, this dissertation employed the technology of non-linear integration and track restriction, which enhanced the energy of target, and the same time, utilized the overlapped characteristic of the slow moving target, as a result, the excellent detection performance was produced.To the moving background, this dissertation addressed the detection of the image target in the complex noisy, and proposed two adaptive filtering technologies. PC experiments indicated that the performance of these technologies was excellent. This dissertation proposed two MF-based integrated detection technologies: one based on the MF clustering; the other based on the MF track restriction. PC experiments revealed that these technologies were not sensitive to estimation precision of motion vector, but only demanded a certain difference between the targets and background motion vector. This feature highly improved the reliability of these technologies.