Video analytics can be used in different applications, including traffic control, security surveillance, and factory floor monitoring. A typical video analytics application consists of a pipeline of video processing modules including an NN-based object detector. The pipeline has several knobs such as frame resolution, frame sampling rate, bitrate, and detector model. The choice of configuration impacts resource consumption, latency, bandwidth requirements, and accuracy of the video application. The best configuration for a video analytics pipeline also varies over time, often at a timescale of minutes or even seconds. In many cases, the policy that reduces the frame rate and lowers the resolution can save resources without impacting the accuracy. In this project, we design and evaluate a controller that dynamically picks the best configurations for existing NN-based video analytics pipelines.
Lab: Wireless and Mobile Networking Lab
Direct Supervisor: Mahshid Ghasemi
Position Dates: Spring May 31-Aug. 19
Hours per Week: 35
Position type: Hybrid (both remote and on site)
Programming languages: C/C++, Python, Shell scripting
Computer vision: previous experience in working with Deep learning models for object detection/tracking e.g. YOLOv4/Nvidia DCF tracker
Preferred: Familiarity with the GStreamer library, DeepStream SDK
Mahshid Ghasemi <[email protected]>