Research Of Fast Tracking Algorithm For Motion Targets Based On GPU

The visual target tracking technique is based on the feature information of moving object in image sequence,and predicts the size and position of the target in the subsequent image frame.It is an important research direction in the field of machine vision and has been widely used in traffic monitoring,visual navigation,human-computer interaction,medical diagnosis and so on.Recently,with the deeper research on the law of flying insects in the field of biology,ecology and biomimetic robotics,the fast visual tracking technology based on moving objects has attracted more and more attention.In practical applications,tracking algorithms for moving targets often require high operating efficiency,low complexity and low computational overhead.However,the traditional visual tracking algorithms generally fail to meet the requirements of fast tracking on tracking speed.This paper focuses on the fast visual tracking algorithm of moving targets.The specific work is as follows:By summarizing and analyzing the existing moving target tracking algorithms,the CAMShift algorithm is chosen as the theoretical basis.The algorithm has less computational overhead and is less affected by illumination.It can realize effective tracking with a single tracking target and non-complex background.Aiming at the characteristics of fast moving target,some improved schemes are proposed for the algorithm.The parallelization method of the algorithm is designed and the improved CAMShift algorithm is parallelized on the heterogeneous platform of CPU and GPU.Its structure is optimized and the exchange of data between CPU and GPU is reduced,which lead to the improvement of tracking efficiency.Effectiveness of the algorithm is verified on the standard tracking dataset VOT2014 Benchmark and the real-time tracking experiment of moving targets is carried out based on the high-speed real-time visual platform.The offline and online experimental results verify that the algorithm can maintain a better tracking accuracy at a high tracking speed and achieve the desired goal of effective fast tracking of moving targets.Given the characteristics of wide moving range,fast moving speed and short reaction time of moving target,the traditional tracking system with a fixed camera position and a small shooting range results in losing moving target easily.A set of 2 degree-of-freedom active vision servo tracking platform is designed and developed,which can control the tracking camera to complete left and right rotation and the pitch motion simultaneously to achieve active tracking according to the position feedback of the tracking target in the image.The system integrates the improved CAMShift algorithm and possesses compact mechanical structure and short system response time to ensure that the tracking target is always in the camera's field of view when it moves in a larger space range.Under the Visual C++ development environment,we implemented the functions of the system based on OpenCV,CUDA and other open source libraries.The tracking experiments are carried out and the experimental results verify that the system enables active tracking of moving targets in a wide range,which performs practical engineering application value.