31
edits
Changes
→Summary
=== Summary ===
Our code changes involved: 1.Parallelized our code with the '''#pragma omp parallel for''' statement. * This improved our runtime to 5(+- 1 second) seconds of OpenCV overehead + length of video for a 1080p video 2. Added dynamic scheduling with our code * Improvement on load balancing. There was minimal runtime gain on our test videos because initial parallelization already optimized the code to handle frames as fast as they are retrieved 3. Eliminated the data type conversion in our logic * Improvement on memory management. There was minimal runtime gain on our test videos because initial parallelization already optimized the code to handle frames as fast as they are retrieved however the section itself showed a 50% improvement in runtime To preface our results: We used OpenCV to process and display our results. This added roughly 5 seconds of overhead added by OpenCV to create the Window and the related processes required to display the resulting video. This is outside of our control. After our initial parallelization improvements our bottleneck is no longer our code but the speed at which we get frames from the source so our improvements do not improve our runtime with a 1080p video but could help maintain this level of execution on higher resolution videos.todo