الفهرس Only 14 pages are availabe for public view
 |  | from | 145 |  |  |
| | | from | 145 | | |
Abstract
Object tracking is vital in our life where it is used in many practical applications such as traffic monitoring, video surveillance, motion-based recognition systems, etc. Object tracking is divided to two main processes namely motion detection and Motion Estimation (ME). Motion detection occurs simply by subtracting successive frames on pixel by pixel basis. The difference image carries information about motion existence. ME is the process of determining motion vectors of a moving object. Block Matching Algorithms BMAs) are the most commonly used motion estimation technique. Its basic idea is dividing the current frame into a matrix of macro blocks that are compared with corresponding blocks in the previous (reference) frame including its adjacent neighbors inside a search window. In this thesis a tracking method using block matching algorithms is presented. The BMA is applied only to Difference technique is first applied on two successive image frames to define ROI where motion occurs. Next, BMA is applied to the defined ROI .The produced motion vectors are scanned and their repetitions are marked. The maximum repetition motion vector is used to represent the motion of the tracked object. This is achieved by grouping the macro blocks that have these repeated motion vectors together. BMAs are grouped to two main categories Full Search BMA (FSBMA) and Fast Block matching algorithms(FBMAs) .To have fast tracking process, fast search algorithms are essential. A main goal of V thisthesis is to develop new (FBMAs) that are less complex than the existing ones. Two proposed FBMAs namely One Point Pentagon Inner Search OPPEN) and Predicted One Point Hexagonal Inner Search (POPHEX) are developed. The proposed algorithms were compared with most recent FBMAs for both motion estimation and object tracking. MATLAB environment is used for development and comparison. The analysis of the comparison results shows the performance gain of the proposed algorithms. Also tracking without using ROI is performed to demonstrate the performance gain of using ROI.Hardware design and implementations of ME algorithms based on FPGA can exploit pipelined and massively parallel processing, resulting in faster and cost effective ME rather than software implementations. A lot of hardware implementations were directed to FSBMAs due to its regularity and high performance. However FSBMA is a very time consuming method which leads to large tracking time. Hence, our goal was to implement FBMA in H/W. Examples from FBMAs are selected such as Diamond Search (DS) and Hexagon Based Search Algorithm (HEXBS).HEXBS algorithm is implemented using Xilinx xc5vlx30-3ff676 device. VHDL is used to describe the algorithm. Modelsim is used for simulation. Xilinx environment is used for synthesis and post route simulation. The results are presented in the thesis.