Bio-inspired visual neural network for stop-and-go wave perception in high-density crowd

In high-density crowd, a unique visual motion effect called <i>stop</i>-<i>and</i>-<i>go wave</i> occurs, which could evolve to trampling and compression incidents. However, few computational models have been reported for stop-and-go wave perception form the angle of computer vision. On the bases of the neural structures of the locusts’ vision systems and the frequency resonance properties of the interneuronal (<i>INs</i>) neurons in locust’s brains, this paper investigates a bio-inspired visual neural network (sgWPNN) for <i>stop</i>-<i>and</i>-<i>go wave</i> perception in high-density crowd scenes. The proposed sgWPNN firstly visually stabilizes the image frames; and then converts the spatial intensity visual information to the frequency domain for extracting crowd’s motion cues; finally, sgWPNN output membrane potential impulses to identify the occurring of stop-and-go wave in the field of view. Videos filmed high-density crowd activities in two scenarios demonstrate the effectiveness of sgWPNN in the perception against to the unique stop-and-go wave effect in moving crowd scenes. This work discusses the biological inspired processing of dynamic visual information in crowd activity perception, which can provide some new ideas for crowd activity detection and behavior analysis.