Research And Implementation Of Visual Target Tracking Algorithm Of Mobile Terminal

With the rapid development of artificial intelligence, along with relevant technology and application of its come into being. AI-based industry is booming.Image recognition and object tracking technology is widely used in various fields,especially with regard to security monitoring and augmented reality. In terms of accuracy, the appearance of the background and objectives in the video image may change at any time. Existing image processing algorithms are difficult to accurately identify the target in the image sequence. In real time, image processing algorithms generally model the appearance of background and objectives in the video image based on the pixels, which makes it have a large amount of calculation features.Especially when faced with a large number of images to be addressed, solving the problem of real-time in target tracking algorithm is significant. With the computing power of the mobile terminal is greatly limited, Real-time target tracking problems in the mobile terminal becomes more prominent. One solution is to complete the target tracking task on the server, such as some mobile augmented reality programs using network to transmit the image to the server. Then the server transmit the image processed by the PC back to the mobile terminal. However, this method is limited to use in a good network environment. It is not general. The purpose of this work is to make the mobile terminal being able to track targets in real time using the computing resources of the mobile terminal itself and ensuring the tracking accuracy of the premise, by upgrading its computing speed.This paper studies the correlation tracking algorithm based on discriminant model. Then make a detailed analysis of the principles and processes of two kinds of tracking algorithm, which are TLD and KCF. The advantages and disadvantages of the both kinds of tracking algorithms are also analyzed. This paper try to retain the advantages of their high tracking accuracy and fast tracking speed and I focus on the algorithm’s real-time and anti-blocking ability. The original TLD concerns more about the detection result of the detector, which weakened the tracker’s role. As a result of the sliding window mode, the detector needs to handle a large number of image blocks, which makes a serious impact on the real-time. In this paper, with the characteristics of high accuracy and timeliness of the KCF, I replace the tracker of TLD. Thus I proposed my new tracking algorithm which named KCF-LD.This paper studies the JNI mechanism of android application development for tracking algorithm containing a large number of digital computing. I use C++language to implement the large number of digital computing by the JNI mechanism.Then Java language can call the method from the compiled DLL. And I test the proposed algorithm which named KCF-LD on the PC side and the experimental results are compared and analyzed, which proves that the KCF-LD not only has anti-blocking properties but also has a improved speed. Then on the Android system I achieve a target tracking system, whose tracking module is based KCF-LD. And I test three scenarios, which proves KCF-LD’s availability in the mobile terminal.