A new crowd anomaly detection model: optimisation aided detection and localisation

Detecting anomalies in crowded scenes is a challenging and important part of the intellectual video supervision system. In this work, a novel crowd anomaly detection model is developed that includes three main phases. Firstly, Entropy-based FCM (EFCM) is carried out to segment the frames from video. Further, Histogram of Gradients (HoG), Local Gradient Pattern (LGP) and improved motion estimation features via block matching technique features are derived. These features are then classified via hybrid classifiers that include DBN and LSTM models. The weights of DBN and LSTM classifiers are tuned in an optimal manner via Ant Lion Aided Grasshopper Optimisation (ALA-GO) model. The suggested HC + ALA-GO model achieved the minimal cost of 0.65 and provides faster convergence rate compared to HC + GOA, HC + SSO, HC + SMO, HC + ALO, HC + SSA and HC + SI-DOX models.