Yapay zekâ ile akilli şehirlerde çevresel görüntü ve video analizienvironmental image and video analysis in smart cities with artificial intelligence

This thesis aims to develop advanced artificial intelligence (AI)-based image and video analysis methods to enhance the quality of life and strengthen smart city security systems. The study focuses on three core problems: fire detection, anomaly detection, and crowd analysis, offering original solutions for each.In the fire detection domain, an innovative histogram-based method grounded in the hue, saturation, value (HSV) color space — which better represents human visual perception of color — has been developed. Extensive experiments on the FLAME dataset demonstrate that the proposed method effectively distinguishes between fire and non-fire scenes, achieving high classification accuracy. Various ML algorithms, including SVM, Decision Tree, and K-Nearest Neighbors, were employed, and their results were compared. We analyzed the motion and color features extracted from the video streams using effective feature extraction techniques, such as the histogram of oriented gradients and color histograms for anomaly detection. Classification using multiple ML models enabled the precise identification of high-sensitivity abnormal events captured by security cameras. Two DL architectures, CSRNet and MCNN, were used to generate density maps and perform crowd counting on the ShanghaiTech and Mall datasets. The experimental results reveal that the CSRNet model provides more accurate and detailed predictions, especially in highly congested and complex scenes, whereas the MCNN model offers computational efficiency and faster inference, making it a viable option for resource-constrained applications. Experiments and analyses demonstrate the reliability, high accuracy, and practical applicability of the proposed methods across various real-world scenarios. This thesis makes significant contributions to the development and implementation of artificial intelligence (AI)-driven image processing techniques to improve security, emergency management, and public monitoring within smart city frameworks. The findings and developed models presented here represent a crucial advancement in smart city technologies and provide a solid foundation for future research in this field.