Loading...
Publication
Shopping list automator: geração automática de listas de compras de supermercado
| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 8.02 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
O ritmo de vida atual é cada vez mais assoberbado por tarefas repetitivas que dificultam muitas vezes a gestão do tempo e da vida pessoal e familiar. Muitas destas tarefas, que são efetuadas rotineiramente, podem no entanto ser otimizadas ou mesmo automatizadas. Falamos em particular da realização de compras de supermercado, que se repetem diária, semanal ou mensalmente, tirando-nos muito tempo e colocando stress na organização do dia-a-dia.
Este trabalho foca-se num sistema capaz de sugerir listas de comprar automáticas, baseando-se no padrão de consumo do utilizador. Tipicamente temos por costume verificar repetidamente que produtos nos faltam em casa, sempre que nos deslocamos às compras. Este sistema procura simplificar este planeamento, sugerindo ao utilizador uma lista que tem em consideração o consumo médio registado de cada produto. A solução desenvolvida baseia-se na utilização de uma aplicação móvel usada para registar o histórico de faturas de compras realizadas. Esta aplicação permite também submeter as faturas a um serviço de backend para registo e análise de consumos. As faturas são importadas através da captura de fotografias ou selecionando imagens ou ficheiros pré-existentes no smartphone. No servidor efetua-se inicialmente a união de imagens de faturas mais longas que não cabem numa única foto, seguindo-se o pré-processamento da imagem, extração de texto e categorização de informação que permite armazenar numa base de dados toda a informação relativa aos produtos adquiridos pelo utilizador. Posteriormente, os dados históricos de consumo do utilizador são usados para estimar as necessidades de produtos e sugerir automaticamente uma lista de compras na próxima ida ao supermercado.
As informações relativas às faturas importadas, respetivos produtos adquiridos e a lista de compras automática são apresentadas na aplicação móvel. O serviços de backend permite identificar o supermercado e a data de cada fatura, os produtos contidos na mesma e a respetiva quantidade, capacidade e preço unitário. Para a extração de texto a partir de imagens das faturas de compras recorreu-se a uma ferramenta de OCR denominada Tesseract, que foi combinada com um componente desenvolvido neste trabalho para interpretar e categorizar a informação recolhida pelo OCR. A avaliação da precisão da extração do texto e da sua categorização em produtos e quantificação de consumo, foi efetuada através de comparações entre os resultados obtidos pelo componente proposto e os valores de teste esperados. Para esta avaliação recorreu-se a métricas de comparação de strings e a um script que analisa e compara a informação recolhida com uma estrutura JSON pré-definida de produtos. Todo o sistema desenvolvido foi descrito nesta dissertação, juntamente com os testes efetuados que permitiram aferir o cumprimento dos requisitos inicialmente propostos.
The current pace of life is increasingly overwhelmed by repetitive tasks that often make it difficult to manage time and personal and family life. Many of these tasks, which are performed routinely, can however be optimized or even automated. We are talking in particular about grocery shopping, which is repeated daily, weekly or monthly, taking a lot of time and putting stress in the organization of our day-to-day activities. This work focuses on a system capable of suggesting automatic shopping lists, based on the user’s consumption pattern. Typically, it is our custom to repeatedly check what products we are lacking at home, whenever we go shopping. This system seeks to simplify this planning, suggesting to the user a list that takes into account the average consumption registered for each product. The developed solution is based on the use of a mobile application used to record the history of purchase invoices made. This application also allows submitting invoices to a backend service for recording and analyzing consumption. Invoices are imported by capturing photos or selecting pre-existing images or files on the smartphone. On the server, images of longer invoices that do not fit into a single photo are initially joined, followed by image pre-processing, text extraction and information categorization that allows all information relating to the products purchased by the user. Subsequently, the user’s historical consumption data is used to estimate product needs and automatically suggest a shopping list on the next trip to the supermarket. Information regarding imported invoices, the respective purchased products and the automatic shopping list are displayed on the mobile application. The backend services make it possible to identify the supermarket and the date of each invoice, the products contained in it and the respective quantity, capacity and unit price. To extract text from images of purchase invoices, an OCR tool called Tesseract was used, which was combined with a component developed in this work to interpret and categorize the information collected by OCR. The evaluation of the accuracy of the text extraction and its categorization into products and consumption quantification was carried out through comparisons between the results obtained by the proposed component and the expected test values. For this evaluation, we used string comparison metrics and a script that analyzes and compares the information collected with a pre-defined JSON structure of products. The entire system developed was described in this dissertation, together with the tests carried out that allowed to assess compliance with the initially proposed requirements.
The current pace of life is increasingly overwhelmed by repetitive tasks that often make it difficult to manage time and personal and family life. Many of these tasks, which are performed routinely, can however be optimized or even automated. We are talking in particular about grocery shopping, which is repeated daily, weekly or monthly, taking a lot of time and putting stress in the organization of our day-to-day activities. This work focuses on a system capable of suggesting automatic shopping lists, based on the user’s consumption pattern. Typically, it is our custom to repeatedly check what products we are lacking at home, whenever we go shopping. This system seeks to simplify this planning, suggesting to the user a list that takes into account the average consumption registered for each product. The developed solution is based on the use of a mobile application used to record the history of purchase invoices made. This application also allows submitting invoices to a backend service for recording and analyzing consumption. Invoices are imported by capturing photos or selecting pre-existing images or files on the smartphone. On the server, images of longer invoices that do not fit into a single photo are initially joined, followed by image pre-processing, text extraction and information categorization that allows all information relating to the products purchased by the user. Subsequently, the user’s historical consumption data is used to estimate product needs and automatically suggest a shopping list on the next trip to the supermarket. Information regarding imported invoices, the respective purchased products and the automatic shopping list are displayed on the mobile application. The backend services make it possible to identify the supermarket and the date of each invoice, the products contained in it and the respective quantity, capacity and unit price. To extract text from images of purchase invoices, an OCR tool called Tesseract was used, which was combined with a component developed in this work to interpret and categorize the information collected by OCR. The evaluation of the accuracy of the text extraction and its categorization into products and consumption quantification was carried out through comparisons between the results obtained by the proposed component and the expected test values. For this evaluation, we used string comparison metrics and a script that analyzes and compares the information collected with a pre-defined JSON structure of products. The entire system developed was described in this dissertation, together with the tests carried out that allowed to assess compliance with the initially proposed requirements.