Varfolomieiev A. The methods, software and hardware of increased performance for object tracking in video sequences

The object of research is the digital image processing during object tracking. The purpose is the increasing of the efficiency of object tracking through the development and improvement of methods and structure of tracking systems, software and hardware for their implementation including implementation on embedded DSP and SoC hardware platforms. The methods of investigation are k-mean clustering, mean-shift, weak classifiers boosting, correlation and differential matching, Lucas-Kanade optical flow estimation. Theoretical results are following: the tracking method based on pyramidal Lucas-Kanade optical flow estimation was improved by the use of adaptive calculation of aperture size and number of levels of pyramid that permits to increase the reliability of tracking of small and low-contrast objects and processing speed; the tracking method based on estimation of density maximum of probability distribution was improved by means of the use of LBP-features and multiscale representation, which allow to increase the reliability of tracking of textured objects on greyscale video sequences and keep the processing speed in the real time; the new method for automatic optimisation of object tracking systems and its Similink-model based on minimisation of object position deviation on reference video sequence and which permits to increase reliability of tracking of systems was proposed and tested; the structure functional organisation of tracking systems was improved by means of introducing the procedure of precise object selection using image segmentation methods. Practical achievements are the creation of algorithms and programs of object tracking systems implemented on PC and on embedded DSP and SoC hardware platforms that able to process video on 25 fps; the suggestion of the new methodology for fast prototyping of tracking systems on DSP platform using MATLAB Simulink; the formulation of recommendations on the feasibility of using DSP, SoC and FPGA hardware platforms for implementation of embedded object tracking systems; the development of prototypes (mockups), which are able to operate in real-time on the evaluation modules based on DSP and SoC with ARM core. The results of the work were adopted in organization Ltd. "Video Internet Technologii", in the learning process of NTUU "KPI" and used in results of R&D in NTUU "KPI". Scope: human-machine interfaces, security and surveillance systems, medicine, driver assistants, robotics, optical location and guidance systems.