A neural network aided attuned scheme for gun detection in video surveillance images

Abstract

Closed Circuit Television (CCTV) cameras are installed and monitored in private and open spaces for security purposes. The video and image footages are used for rapid actions, identity, and object detection in commercial and residential security. Object and human detection require different classifications based on the features exhibited from the static/ mobile footages. This article introduces an Attuned Object Detection Scheme (AODS) for harmful object detection from CCTV inputs. The proposed scheme relies on a convolution neural network (CNN) for object detection and classification. The classification is performed based on the Object's features extracted and analyzed using CNN. The hidden layer processes are split into different feature-constraint-based analyses for identifying the Object. In the classification process, feature attenuation between the dimensional representation and segmented input is performed. Based on this process, the input is classified for hazardous objects detection. The consecutive processing layer of CNN identifies deviations in dimensional feature representation, preventing multi-object errors. The proposed scheme's performance is verified using the metrics accuracy, precision, and F1-Score.External dataset training has improved accuracy by 8.08% and reduced error and complexity by 7.47 and 8.23 percentage points, respectively, in this process. Object classification based on labels is expected to be implemented in the future.