Intelligent Detection Of Crime Anomalies In Smart Cities Using Hybrid Machine Learning With Improved Segmentation And Feature Extraction Techniques

Abstract

The rise in population in urban areas has resulted in difficulties in policing and monitoring high-crime probability areas, leading to an increase in criminal activity and insecurity. To enhance security, smart cities have integrated crime detection systems with video surveillance as the standard method. The backlog of video data that must be monitored by supervising officials can lead to an increase in error rates. To address this issue, a proposed solution involves using meta-heuristic optimization with a Hybrid Machine Learning algorithm. This solution analyzes video stream data quickly and accurately, facilitating the identification of criminal activity. This approach is expected to improve the efficiency and effectiveness of video surveillance systems. The proposed method involves pre-processing the video data using techniques such as Video-to-Frame Conversion, Resizing, and Normalization, followed by segmentation of the frames using an optimized Semantic Segmentation- Optimized FCN algorithm. Features are then extracted from the segmented regions using techniques such as SIFT and the proposed Improved Histogram of Oriented Gradients algorithm. The extracted features are refined using the new improved Relief Algorithm for feature selection. Lastly, a new hybrid machine learning approach is designed using a combination of transformer model, SVM, and ANN for crime anomaly detection. The proposed method is implemented using the Python programming language.