Title

Dynamically parallel CAMSHIFT: GPU accelerated object tracking in digital video

Date of Award

2016

Document Type

Thesis

Degree Name

Master of Science (MS) in Computer Science

Department

Computer Science

Abstract

"The CAMSHIFT algorithm is widely used for tracking dynamically sized and positioned objects in real-time applications. In spite of its extensive study on the platform of sequential CPU, its research on massively parallel Graphical Processing Unit (GPU) platform is quite limited. In this work, we designed and implemented two different parallel algorithms for CAMSHIFT using CUDA. The first design performs calculations on the GPU, but requires iterative data transfers back to the host CPU for condition checking, which bottlenecks the entire program. In the second design, we propose an enhanced parallel reduction-based CAMSHIFT using dynamic parallelism to reduce overhead of data transfers between the CPU and GPU. Test results for a 400 by 400 search window show that the second design is up to five times faster than the first design and nine times faster than a pure CPU implementation. We also investigate the deployment of dynamic parallelism for multiple object tracking using CAMSHIFT"--Leaf iv.

Creative Commons License

Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.

This document is currently not available here.

Share

COinS