Object-tracking is a fundamental problem in the field of intelligent video,has been the academic concerns and have constant progress.Meanwhile,with the development of embedded technology,the performance of embedded micro processors is much higher than before,which made it possible that the object-tracking algorithm can be planted in embedded system.This dissertation firstly analyses the recent object-tracking algorithms and proposes an algorithm with less calculation based on the key hue to meet the need of embedded-platform.The key Hue is an interval of hue,which can present the target in certain range of the image.We convert the image to HSV space,then analyze the hue distribution of target and background in the initial image and get the key hue interval.In the following image,we use the key hue interval as threshold to convert the image to binary format and track the object according to the change of centroid of the binary image.Experiments results in the open video set show that the algorithm has a much fast speed than recent mainline ones.On this basis,we propose two methods to enhance the precision of tracking.One is using adaptive interval threshold width,another is adding window function on the patch to process.The main purpose of both methods is to reduce the noise.At last,we transplant the object-tracking algorithm based on key hue to embedded platform and verify the feasibility of the object-tracking algorithm in the embedded system as a form of android application. |