| Summary: | 碩士 === 國立東華大學 === 資訊工程學系 === 97 === This thesis aims at developing machine vision techniques for detecting vehicles from side views using neural networks. The edge-based SIFT distinctive feature detector is proposed to find more accurate and robust feature points comparing to traditional SIFT feature detector. Further, a two stage subspace-based filter is exploited to filter out the outliers and pick up the significant features. STD filter is used to filter out the outliers, whereas PCA filter is used to pick up the significant features. In doing these filtering, not only the computation time speeds up but also the distinctive features locate. Finally, the Backpropagation Neural Network is employed to detect the vehicles from side views. The experimental results show that the proposed method is capability to deal with two major detection problems, scaling and partial occlusion, under complex background.
|
|---|
