Keywords
ITL (KT0), Visión artificial, Códigos de barras, Códigos QR, Drones, Inventario., Computer vision, Bar codes, QR codes, Drones, Inventory.
Sistema de inventariado para drones usando visión artificial
Abstract
La realización del inventario es una tarea básica en el proceso de producción de cualquier almacén que, sin embargo, puede resultar bastante tediosa si se realiza de forma manual. Los drones se presentan como una posible solución a la automatización del inventariado ya que permiten registrar paquetes que se encuentran en zonas de difícil acceso. Para realizar la extracción de la información de cada paquete podría instalarse un lector de códigos de barras y QR en el dron, sin embargo, dicho sensor eleva el consumo de batería del dispositivo reduciendo su autonomía y aumenta el peso del conjunto dificultando el vuelo. Por ello, se plantea el uso de visión artificial para automatizar el proceso de inventariado, partiendo de imágenes de los paquetes tomadas en tiempo real por la cámara que suele estar ya instalada en el dron. El presente proyecto describe un algoritmo que permite localizar y leer códigos de barras y QR a partir de imágenes tomadas con diferentes orientaciones y condiciones de iluminación. Dicho algoritmo ha sido implementado sobre la Raspberry Pi Zero W. El algoritmo está compuesto por tres fases ejecutadas de manera secuencial: preprocesamiento (elimina toda la información de la imagen que no es necesaria para la lectura del código haciendo el archivo más ligero para fases posteriores), localización (detecta la posición exacta del código dentro de la imagen) y lectura (extrae la información de dicho código). Para la implementación del algoritmo se ha hecho uso de la librería de tratamiento de imágenes OpenCV y la librería de lectura de códigos ZBar. El algoritmo diseñado consigue detectar y decodificar más del 80% de los códigos con distintas condiciones de iluminación y orientaciones, empleando una media de 787,00 milisegundos en la Raspberry, tiempo que no compromete el funcionamiento del dron.
Carrying out the inventory is a basic task in the production process of any warehouse that, however, can be quite tedious if done manually. The drones are presented as a possible solution to the automation of the inventory because they enable to register packages placed in areas of difficult access. To extract the information of each package, a barcode and QR reader could be installed in the drone, however, this sensor increases the battery consumption of the device, reducing its autonomy and increasing the weight of the assembly, making it difficult to fly. Therefore, the use of artificial vision is proposed to automate the inventory process, based on images of the packages taken in real time by the camera that is usually already installed in the drone. The present project describes an algorithm that allows locating and reading barcodes and QRs from images taken with different orientations and lighting conditions. This algorithm has been implemented on the Raspberry Pi Zero W. The algorithm is composed of three phases executed sequentially: preprocessing (eliminates all image information that is not necessary for reading the code making the file lighter for later phases), location (detects the exact position of the code within the image) and reading (extracts the information from that code). For the implementation of the algorithm, the OpenCV image processing library and the ZBar code reading library have been used. The designed algorithm manages to detect and decode more than 80% of the codes with different lighting conditions and orientations, using an average of 787.00 milliseconds in the Raspberry.
Carrying out the inventory is a basic task in the production process of any warehouse that, however, can be quite tedious if done manually. The drones are presented as a possible solution to the automation of the inventory because they enable to register packages placed in areas of difficult access. To extract the information of each package, a barcode and QR reader could be installed in the drone, however, this sensor increases the battery consumption of the device, reducing its autonomy and increasing the weight of the assembly, making it difficult to fly. Therefore, the use of artificial vision is proposed to automate the inventory process, based on images of the packages taken in real time by the camera that is usually already installed in the drone. The present project describes an algorithm that allows locating and reading barcodes and QRs from images taken with different orientations and lighting conditions. This algorithm has been implemented on the Raspberry Pi Zero W. The algorithm is composed of three phases executed sequentially: preprocessing (eliminates all image information that is not necessary for reading the code making the file lighter for later phases), location (detects the exact position of the code within the image) and reading (extracts the information from that code). For the implementation of the algorithm, the OpenCV image processing library and the ZBar code reading library have been used. The designed algorithm manages to detect and decode more than 80% of the codes with different lighting conditions and orientations, using an average of 787.00 milliseconds in the Raspberry.
Información detallada
Materias, derechos, colecciones e identificadores