Pedestrian Detection Based On Android Platform

In recent decades,with the progress of society,computer technologies have got rapid development.Nowadays,computer intelligent visual technology has become a popular research direction,and in particular,pedestrian detection technique attracts more attention from researchers.Pedestrian detection technology can be applied in a very wide range of applications.Such as vehicle driver assistance system that devotes itself to improve traffic safety,population forecast and management system,and so on.Pedestrian detection also has high application value in the robot and advanced human-computer interaction.On the other hand,because of the advantages of mobile devices including small size,easy to carry and high performance,mobile electronic products attract a large number of users.The application market of Android operating system for mobile terminals also gets fast popularization.Therefore,it is not only necessary but also significance to transplant the pedestrian detection technology to the Android platform.Transplant of the OpenCV based pedestrian detection technology to the Android platform makes users processing the image and video more flexibility and mobility.In addition,it also broadens the scope of application of the Android platform.Since pedestrian detection has broad application areas and the Android platform has good prospects for development,this work focuses on implementing pedestrian detection for Android mobile platforms.The contents of this paper consist of four main sections1.We briefly introduce the history of Android operating system,including the birth,development and prosperity of the Android system,and describe in detail the composition and structure of the Android system.The life cycle,four main components and the application process are also introduced in this section.2.Describing characteristics of pedestrian detection technology including SIFT feature,HOG feature and LBP feature.The traditional training algorithm for pedestrian detection is studied,such as SVM algorithm.In addition,this paper studies the whole process of training cascade classifier using Adaboost framework.After that,we studied the rectangular box generated by integrating multiple detection windows based on multi-scale image utilization and check-up technique.3.The INRIA dataset is used as the training set and test set for experiment.In addition,the test set also includes images took from a mobile phone and local images.Considering the limitation of HOG feature and the LBP feature has some supplementary on its performance,the LBP-HOG joint feature is selected as a descriptor for characterizing pedestrians.Because of the increase of the dimension of the joint feature,pedestrian detection using the SVM classifier is a time consuming task on mobile devices.Therefore,the Adaboost algorithm is adopted to train the cascade classifier in order to reduce the complexity of the algorithm and increase the accuracy of pedestrian detection.4.We build experimental environments on both personal computers and mobile platforms.At first we build the Android system environment and transplant the OpenCV library to the Android platform.Then the pedestrian detection programs developed on personal computers are modified to implement the proposed mobile platform based pedestrian detection system.