Enhanced Public Security: Modified YOLO Network for Unattended Object Detection in Dynamic Environments

The majority of people will automatically assume that something is amiss if they discover any item left unattended in a public area. It is one of the trickiest computer vision difficulties to identify such objects in a broad environment. Public spaces like malls, airports, and train stations can detect safety violations more precisely and swiftly with robust object detection systems. This research proposes the Modified YOLO Network as a novel and effective method for single-camera unattended event detection. Using the background removal methodology, the proposed method separates objects from the foreground. You can further categorize these foreground items as moving or stationary by looking at their centroid positions in successive frames. Foreground objects are considered neglected if they remain motionless for a certain amount of time and do not grow larger than a certain threshold. In order to extract multi-dimensional properties, we use the BotNet attention mechanism to build a modified YOLO detection algorithm. Because of this, the model is able to quickly converge with a small sample set and extract features from small targets more effectively. Lastly, Shape_IOU is used to compute the training loss. Taking the bounding box's form regression loss into account also aids in improving the representation. To improve the classification accuracy, the parameters of the upgraded YOLO network are adjusted with the aid of the altered chaotic grasshopper optimiser. One can evaluate the efficacy of the proposed strategy by testing it on the PETS2006 dataset. To reduce the number of false alarms caused by a dynamic background, you can change or replace the first backdrop frame. It is effective for object identification in various public settings and is generic.