Date of Award
Master of Science (MS)
Electrical and Computer Engineering
This thesis proposes to work on a parallelization method to speed up the computational runtime of the particle filter algorithm for multiple targets tracking. CUDA programming is utilized to execute the original implementation of the particle filter algorithm on GPU. The thesis provides a detailed discussion of the background information on the relevant topics. And then a presentation of the code architecture changes is followed. The detailed CUDA-based implementation is illustrated and discussed, which is followed by a discussion andcomparison of the results obtained from a series of tests.In this thesis, the introduction and description of the basic particle filter are presented first. Detailed illustrations of each step in the original implementation of the particle filter algorithm, which is executed sequentially on CPU, are provided. Then, background information of parallel programming technologies is provided, such as GPGPU and CUDA programming. The new design of the CUDA based implementation of the particle filter algorithm is proposed to speed up the execution of the original implementation, which is executed on CPU. Moreover, a detailed explanation of the CUDA-based implementation is given.Finally, the thesis will demonstrate the test results for both CPU and CUDA implementation as a comparison. The experiments indicate that the CUDA implementation can obtain a maximum of 7.5x speedup over the original implementation. After implementing more results and comparison, it was concluded that the CUDA implementation was significantly faster than the CPU version. Furthermore, the CUDA version still has much space for future optimizations to increase its performance.