Crowd visualization on low bandwidth mobile devices based on video analysis
Shuren Tan, Xuan Chen, Yue Yin, Justin Wu, Shenshu Zhou, Zhihui Xiong, Anup Basu · 2016
Many applications need the transmission of video, which often require large bandwidth in order to obtain the desired details. However, not every detail is important in every application, e.g., surveillance only needs to capture the target of interest and alert the operator at the critical moment when action is necessary. This study aims to save bandwidth and enable many applications on resource constrained mobile devices, by simulating crowd scenes without transmitting the actual video of a scene. Current research in this area assumes the availability of Motion Capture (MoCap) data to facilitate this process. However, this is unrealistic for videos of real-world scenes captured by cameras, since they only provide a sequence of frames. The goal of this work is to emulate marker-based motion capture on a large group of people by accurately tracking pedestrians and simulating their motions in a 3D environment. The tracked positions can be input into a simulation environment to approximate motions of pedestrian movement. While the general scene is simulated realistically in the background, the target pedestrian trajectory can be tracked and monitored accurately. We describe pedestrian tracking using our own panoramic tracking hardware, as well as using normal video cameras. Our results show that it is possible to detect individuals in a crowd and simulate real world crowds from video.