National Participation in Research Project Accurate Surveillance Object Detection based on Visual Consistency ConstraintConstructed and labeled a large video data set for falling down people and/or pedestrians;Embedded Aggregated Channel Features detection algorithm into an existing tracking platform to substitute the manually detection for accurate and real-time automatic pedestrian detection.Bachelor’s Thesis: Pedestrian Detection and Tracking SoftwareDecreased False Detection Rate under 5% by joint decision of Detection and Tracking Model;Extracted Motion-History Image from the video dataset and matched templates with Mahalanobis Distance/Support Vector Machines to realize 80% accuracy for action recognition for test samples;