Supermarket Shelves Regional Segmentation And Commodities Identification Study

Supermarket is a place closely associated with people’s lives. The commodity is the core of the supermarket, and its status is a direct reflection of the supermarket operating information. So if we can acquire the current status of various commodities, this will help to know the supermarket sales and can deal with it timely. The current way of supermarket commodity information acquisition is based on software which can be used for the storage and query of the commodity category and numbers mostly, also a lot of staff are needed to provide supplementary information. There are some shortcomings of this kind of operation: a large demand for labor, not timely information updating and so on. Therefore, there is need for a more intelligent method of analysis of the supermarket goods to provide a more intuitive and timely commodity information.According to these shortcomings, we provide a new supermarket commodity recognition method based on digital image processing technology in this paper. The basic information, such as, category, quantity and location of all goods can be acquired finally. The multimodal characteristic of the vertical projection histogram of the shelf image is used for shelf layering. After that, horizontal projection histograms of multiple color components are combined to obtain the image category split lines. Next, shape and color characteristics are used for SVM classifier recognition to get certain areas labels. Finally, based on the histogram of binary image, single commodity can be segmented based on SKU (Supermarket Keeping Unit).Different sets of samples and color feature extraction method are selected to perform three sets of experiments. The testing images comes from supermarket beverage area, comparative analysis is conducted for recognition time consumption and the recognition rate of the goods. The results show that the3D RGB characteristic takes more time than RGB characteristic, however, the former recognition rate is significantly higher than the latter, and less affected by the size of the samples, commodity recognition result is better.