Classification and Tracking of Vehicles and Pedestrian in Video

• Main Authors: Yu-Sheng Lin, 林育生
• Other Authors: Yuan-Kai Wang
• Format: Others
• Language: zh-TW
• Published: 2004
• Online Access: http://ndltd.ncl.edu.tw/handle/19312327184110010388

碩士 === 輔仁大學 === 電子工程學系 === 92 === Moving object tracking has been an important research topic in numerous computer vision applications. Most papers adopt a prediction-based approach that uses linear block matching and Kalman filter. In this paper, an effective tracking approach using object’s features is proposed. Two important kinds of moving objects: pedestrian and vehicles, are studied in our approach. Exact object’s shape is extracted first in order to obtain accurate features of the object. Three preprocessing steps are proposed to detect the exact object’s shape. Five discriminate features are then extracted for each object. Detail analysis of the five features for the classification of object is described. A tracking scheme by classifying the features with a back-propagation neural network is invented. The proposed approach is verified through several image sequences. 1343 objects, including 469 vehicles and 874 pedestrians, can be effectively tracked in our experiments.