Diseño e implementación de minidrones (uavs): inteligencia artificial aplicada en minidrones enfocado al campo de visión computacional.

Abstract

A sub-branch of the field at artificial intelligence is computer vision, which provides mobile robots the ability to capture images, process them and take decisions. By implementing them in UAVs, different tasks have been worked on, such as supervision of products in large warehouses, correct use of masks outdoors in times of COVID, military activities, etc., The use of automatic and deep learning techniques "machine learning - Deep learning", has become popular with the existence of small computers that have high processing capabilities such as raspberry, NVIDIA and others. In this work, the implementation of autonomous artificial vision algorithms is proposed for the development of two applications: the first consists of monitoring vehicles that circulate through a region of interest "ROI" drawing in some area of a busy road. Thus, counting cars and capturing them for later analysis, the second application consists of detecting and tracking a particular target, which is a person wearing a red jacket. For both applications, video and various images have been taken as a basis, simulating captures taking by a mini - drone, to which additional processing has been carried out, using object detection algorithm techniques, being the pioneer Yolo v5 in its Tiny version and algorithms of robust tracking for multiple objects, including "Deep Sort", which have been chosen considering a future execution in an SBC, currently evaluated in different real applications.