Dynamic Human Target Detection Based On Multi-sensor In Long Range

For detecting moving human target in the open country or an airport and so on, we need full out judge it human beings or not. This paper puts forward a sort of technique of dynamic human target detection based on multi-sensor in long range. Firstly, we introduce the method of background estimation based on kalman filter aiming at dynamic target detection at fixed background in long range, besides improve on the method by changing parameter automatically, in order to detect moving region in infrared image sequence. At the same time, we detect moving region by the idea of optical flow in visible light image sequence. Secondly, we introduce a texture analysis method of grey-primitive cooccurrence matrix aiming at the fact what the amount of pels of human target in long range in picture has between small target and target in close quarters. We gained eigenvectors of moving human target by grey-primitive cooccurrence matrix in infrared image sequence. We gained eigenvectors of moving human target of hominine shape and moving characteristic based on the idea of optical flow in visible light image sequence. Finally, we normalized eigenvectors in infrared image sequence and in visible light image sequence, following we fuse the eigenvectors in multi-sensor. We separate human targets from others through the method of support vector machines after mapping the eigenvectors to high dimensional space in infrared image sequence, in visible light image sequence and fused eigenvectors. The experiment result shows support vector machines is available and feasible method. Comparing the detection result of different conditions, the method of information fusion in multi-sensor is inevitable and exceeds the others, and this technique performs well for dynamic human target detection in long range both in robustness and validity.