Light Weight Filter for Video Object Detection and Visualization Framework for Accuracy Comparison
Author: Farhan Haider
Faculty Supervisor: Shahrukh Humayoun
Department: Computer Science
Object detection on an image is a very compute-intensive process. Running object detection on videos is even more expensive as a video is composed of a huge number of images. In addition to the object detection cost itself, there is also a cost to decode the video. Because of this, running video object detection jobs on low-compute devices is pretty much impractical. This study proposes an efficient approach to video object detection on low-compute devices by only analyzing specific points of the video instead of the whole video. To compare the accuracy of this new system of object detection, a novel visualization framework is introduced. The study shows that this new object detection method can significantly improve the efficiency of object detection without compromising much on accuracy.