| Abstract |
Automated crime anomaly detection systems have been made possible by the influx of urban data streams brought on by the come up of smart cities. The use of machine learning to analyse and spot unusual patterns in criminal incidents is explored in this research. This system improves proactive law enforcement strategies, supports resource allocation, and aids in the development of safer and more secure urban environments by utilising real-time data from numerous sources.The first step is to gather video data from the network of security cameras that have been carefully placed throughout the smart city. With the help of this sizable video dataset, the automated crime anomaly detection system can be trained and improved so that it can learn and distinguish between typical and abnormal patterns of behaviour in a variety of urban settings. From the collected data, preprocessing is processed through Video-to-Frame Conversion, Non-Local Means (NLM) and contrast stretching approach. Sobel edge detection approach is used to Identify the Regions of Interest (ROI) in the frames for the Segmentation from the pre-processed data. To Extract features from the segmented regions, Improved Gradient Local Binary Patterns, Haralick and Gradient Interpolation-Based Hog Model is used. Refine the extracted features to remove any irrelevant or redundant features using the new hybrid optimization approach-Sharkprey Optimization Algorithm that combines the White Shark Optimizer and Osprey optimization algorithm. From the selected features, design a new ensembled-machine learning approach for crime anomaly detection by combining the K-Nearest Neighbors, Random Forest and optimized Artificial Neural Network. Tostrengthen the detection accuracy, the weight of ANN is fine-tuned by the Sharkprey Optimization. MATLAB is used for the implementation. © 2024, Collegium Basilea. All rights reserved. |
