Train and Employ a Faster-RCNN framework to perform pedestrain detection in videos.
Re-train Faster-RCNN on VOC 2007 and VOC 2012, and also Caltech pedestrian dataset to perform pedestrian detection.
Apply the model on videos to generate pedestrian bounding-boxes frame-by-frame. Frame-wise bounding boxes are smoothed by moving average. Then the processed frames were concatenated to generate a new video.
When fed with a video where pedestrians appear during specific frames, the network will process and output a new video with pedestrians marked by the bouding boxes.