OpenCV code to anonymize video

- Project requirement is to create OpenCV code to produce output the same or better as the provided example - It includes both car/pedestrian detection (see below 'inital frames') as well as motion removal - Input images need to be anonymized, one way to do this is all fast moving motion is to be removed, ideally it would be just people and cars that are removed but since that is a harder problem computationally we settle for motion removal as shown in the example video (inital frames however do need people and cars blurred, as defined below 'inital frames' section) - Motion like clouds, shadows, the crane needs to be preserved - Performance is of most importance, will be ran on a server processing many instances in parallel - Assumption is that the camera is not moving, and frames are at least 1sec apart, often they are 1 min to 1 hour apart - Input images can be up to 8K in resolution, processing can be done on a smaller scaled image for speed, however background replacement needs to be done from on original resolution to maintain quality - Use OpenCV 4.1 and Emgu CV, the code will be called from a C# .Net app - Need to be able to initialize the process based on prior state and resume, i.e. process one frame at a time, that is input state + new frame => output frame - Takes in memory BGR frames as input and as output - Old state if any + single input frame -> single output frame + new state - Old state if any + video -> output video - Along with the code, a standalone Windows exe it to be produced that takes mp4 video file as input and produces an mp4 video as output Initial frames - Car/people detection: is to treat initial frames until the background is learned for proper motion removal above, this requires people/car detection on these initial frames, this it is more computationally expensive so done only on the initial frames - Ideally we can choose to remove people and cars separately - Needs to be done using an efficient algorithm - For frames where background isn't known, the detected objects are to be blurred also the coordinates for these detected objects are to be returned