Matching uses cosine similarity over the uploaded Features_CSV vectors, keeps the detected object classes intact, maps person/auto/car/bus detections to the movement-tracking CSV classes, and estimates speed assuming 5km between devices.

This view uses only the new movement-tracking CSV, groups points by IP address, draws each path on the map, and shows the mapped vehicle type and movement type for every recorded point.