Explainable Computer Vision for Scene Analysis to Detect Crime

As the traditional surveillance methods often struggle to accurately decipher objects and activities in complex scenes. This study focuses on incorporating hyperspectral imaging technology in the detection of criminal activities in surveillance systems. Hyper-spectral imaging, with its ability to capture information across numerous spectral bands, offers a solution to detect objects from complex scenes. In this study, two datasets were utilized, one dataset contains RBG images and another contains hyperpectral images. The proposed model is evaluated using metrics like Precision, Recall and F1 score and the results reveal that there is a slightly higher performance when the hyper-spectral data was integrated into the proposed model as it particularly performs well in detecting objects relevant to criminal activities, such as criminals and weapons identification. By utilizing separate validation and test datasets, the proposed model exhibits high accuracy indicating it’s effectiveness in real-world crime detection scenarios.