Architecture for accelerating video compression by using the motion vectors produced locally by a camera is disclosed. Video frames are captured by the camera (e.g., a webcam) which also computes a motion vector for the frames. Metadata can also be generated that represent an index of motion quality associated with the motion vector. The motion vector is passed to a video compression engine which selectively uses the motion vector directly or alternatively as a seed for a compression and encoding algorithm. This algorithm produces a compressed video frame representing a motion estimate having a selected motion quality index value. In this way, complexity is reduced in the video compression engine, resulting in faster and more efficient video compression. Alternatively, the webcam sends a compressed video bitstream to reduce throughput on the connection and the receiving computing system processes residual information to derive an estimate of the quality index for each macroblock/kernel.