监控图像中利用视角分类方法实现目标提取.ppt

5 Future works More Experiments ImageNet Large Scale Visual Recognition Challenge Tofu knife Dragon Falchion Objective function How about “embedded positive space” Do need different classifiers? Do need so many vision tasks? Kindly give your advice. Thank you! Object Detection with Perspective Classifiers 视角 Contents 1 What is Object Detection 2 Results 3 Introduction 4 Proposed Method 5 Future works 1 What is Object Detection Detecting pedestrians from static image 选题意义 人体检测应用领域广泛 智能设备 视频监控 生物特征识别 图像内容理解 基于图像的快速三维建模 1 What is Object Detection Process Learn classifier Encode images into feature vectors/Construct Model Create normalised training data set Human/Non-Human decision Fuse multiple detections in 3-D position scale space Run classifier to obtain object/non-object decisions Scan image at all scales and locations Human detections with bounding boxes Learning Phase Detection Phase 1 What is Object Detection Process Learn classifier Encode images into feature vectors/Construct Model Create normalised training data set Human/Non-Human decision Fuse multiple detections in 3-D position scale space Run classifier to obtain object/non-object decisions Scan image at all scales and locations Human detections with bounding boxes Learning Phase Detection Phase 2 Results (a) Low resolution Qualitative results on Caltech Pedestrian Detection Benchmark Reference papers 2 Results (a) Low resolution (b) Reasonable Qualitative results on Caltech Pedestrian Detection Benchmark 2 Results MT-DPM+Context HOG + Context Multi-resolution Windows ACF + SDt ACF + Motion Information Fix Window MPO (in Experiment) ACF Fix Window 64*32 ; 54*27 MPOL outperforms all other methods in 12 out of all 18 sub-experiments. MPO achieves state-of-art results in 13 out of 18. 3 Introduction Sample (rigid) or Intricate model? Adaboost, Support Vector Machine Deformable Part based Model (DPM) 3 Introduction Sample or Intricate model? Adaboost, Support Vector Machine Deformable Part based Model