Simulating crowd motion using density estimation and optical flow
Di Chen, Gary Kah Meng Tan, Antoine Fagette, Stephen Kheh Chew Chai · 2017
Crowd simulation is often used as a crucial tool to analyse crowd behaviours. Ideally, when analysing live video streams, we would like the simulator to be able to run concurrently. However, crowd video analytics algorithms are usually not able to supply position updates in real time as there exists a noticeable time gap between two consecutive human position updates. the crucial problem is therefore on how to simulate human positions within the time gap. In this paper, a simulation framework that could approximate human displacements in a near real time manner is proposed. A framework based on OpenCV that reads video streams and runs real time simulation is implemented. As a result, amongst the crowd being tracked, we obtain near real time simulation with acceptable tracking accuracy. Lastly, this paper explains the limitation of the proposed framework.