Pedestrian Behavior Recognition via a Smart Graph-based Optimization

Innovative technology and improvements in intelligent machinery, transport networks, emergency mechanisms, and educational facilities define the modern world. It is difficult to comprehend the settings, do crowd monitoring, and observe people. In this research, we present an effective method for understanding crowd data across a sustainable framework and for identifying pedestrian anomalous and normal behavior. Preprocessing is conducted initially, followed by person detection and contour separation. Then, the features are extracted using a force flow matrix and deep flow, and the graph mining algorithm is used to optimize the feature space. Finally, we utilized CNN to distinguish between regular and irregular pedestrian behavior. For the ADOC and UMN collections, the suggested technique has an acceptable performance percentage of 83.30 and 83.83 percent, correspondingly. In conclusion, the acquired results are shown to be more effective than other approaches described in the scientific literature.