Person detection and re-identification across multiple images and videos obtained via crowdsourcing

Person re-identification is indispensable for consistent labeling across different camera views. Most existing studies use static cameras, apply background subtraction to detect moving people, and then focus on the matching of detection results. However, if cameras are mobile or only single image frames (not videos) are available, then background subtraction cannot be used, and human detection needs to be performed on entire images. In this paper, different from most of the existing work, we focus on a crowdsourcing scenario to find and follow person(s) of interest in the collected images/videos. We propose a novel approach combining R-CNN based person detection with the GPU implementation of color histogram and SURF- based re-identification. Moreover, GeoTags are extracted from the EXIF data of videos captured by smart phones, and are displayed on a map together with the time-stamps. All the processing is performed on a GPU, and the average processing time is 5 ms per frame.