Machine inferencing pipeline

Our object detection inferencing pipeline begins with capturing a frame from the incoming video stream. The video stream can be a prerecorded video file, an RTSP stream, or can come from an attached video camera. The frame is then passed to an ML object detector that applies the ML model on the frame to detect a set of objects that the ML model has been trained for. The detector output is a set of detected objects with their bounding boxes, usually a set of coordinates representing a rectangle. This detector output is used to annotate the frame with labels and draw a rectangle around the detected objects. In the interest of keeping the example simple, the annotated frame is then streamed as a Motion JPEG (MJPEG) file so that it can be easily viewed in a browser for verification of the detected objects. It’s possible to stream the detected objects with their bounding boxes to other services or to the Cloud. By locally processing the video stream,...