Humans attributes extraction and search with a deep learning based real-time video analysis system

We present a real-time computer vision demonstration with two main features. Firstly, attendees can visualize the detected person on a camera with a set of estimated attributes such as the clothes color or the clothes length. Then, a search engine enables the participants to request past detections by criteria or photo. The search functionality uses a re-identification system based on a full length human picture, this means we use the global appearance of a person (from head to toe) to decide if two people are similar or not.