Integrating Explainable Artificial Intelligence With Advanced Deep Learning Model for Crowd Density Estimation in Real-World Surveillance Systems

Crowd Density Detection in Smart Video Surveillance involves advanced computer vision (CV) techniques to improve the efficiency and accuracy of crowd monitoring. The system assists in detecting and analyzing crowd density in real-time by utilizing artificial intelligence and machine learning (ML) models on surveillance videos. It detects crowded areas, manages crowd flow, and combines automated analysis with human oversight for improved public safety and early intervention. Explainable Artificial Intelligence (XAI) improves the interpretability and transparency of crowd management methods. Incorporating XAI models provides clear, understandable insights into predictions, ensuring more actionable and reliable crowd management. This study proposes an Osprey Optimization Algorithm with Deep Learning Assisted Crowd Density Detection and Classification (OOADL-CDDC) technique for smart video surveillance systems. The aim of the OOADL-CDDC technique is to enable the automated and efficient detection of distinct kinds of crowd densities. To achieve this, the OOADL-CDDC technique primarily utilizes a bilateral filtering (BF) approach for noise removal process. The OOADL-CDDC technique utilizes an advanced DL method, employing the SE-DenseNet model for feature extraction, while the hyperparameter selection is performed by using the OOA model. Finally, the detection and classification of the crowd density is accomplished by using the attention bidirectional gated recurrent unit (ABiGRU) model. A series of experiments are performed to demonstrate the improved performance of the OOADL-CDDC method. The performance validation of the OOADL-CDDC technique portrayed a superior accuracy value of 98.30% over existing models in terms of distinct measures.