Backpack detection model using multi-scale superpixel and body-part segmentation

Abstract A backpack is a type of carried object (CO) widely used for various purposes because its practicality. Various valuable items such as wallets, laptops, cameras, and cellphones may be kept in backpacks. Detecting backpacks video surveillance challenging due to their varying shapes, sizes, colors. The process localizing the area image critical stage dramatically influences success detection. This paper focuses on through multi-scale segmentation approach, where different scales are intended detect size Based assumption that generally located above bend line, body-part method then select superpixels. selected superpixel feature extracted train model. Model testing out two scenarios. In first scenario, model tested using HOG (histogram oriented gradients) feature, while second combination histogram features. experiment results show DIKE20 dataset, proposed obtained an average F1 score 69%. On PETS2006 i-LIDS datasets, shows 68%, better than by state-of-the-art method.