VIPro-TEC: Desarrollo de una Tecnología basada en Procesamiento de Video e Imágenes Orientada Para Vehículos Autónomos Bajo Condiciones de Visión No Ideales

Technology oriented towards autonomous vehicles has presented an unprecedented advance in
the last decade thanks to the development of complex processing algorithms and the introduction
of high-performance hardware dedicated to these structures. Costa Rica is not exempt from this
type of technology with applications not only in the road field, but also in agriculture, land
exploration, infrastructure, and other fields. These devices have a wide range of sensors to monitor
their environment and use this information for analysis, processing, and decision making in
transport, agricultural, aquatic, aquatic, submarine, war, and other fields. An autonomous air or land
vehicle (such as those proposed in this proposal) use LiDAR (Light Detection and Ranging) type
sensors. the massive data generated by this, and other sensors are usually processed by means
of deep learning methods, which have a high-performance requirement due artificial intelligence
and neural network used for the system development. To reduce this issue, autonomous vehicle
systems would benefit from processing options to support LiDAR sensor failure, alleviating
dependence on a single sensor type to obtain the depth of objects and the environment (even under
less-than-ideal conditions). The work proposed in this project proposes an alternative to LiDAR
technology, presenting an alternative algorithm to make depth processing. The proposed
technology in this Project is based on 3 main aspects: development of the algorithm to calculate
monocular through the use of machine learning and classical image processing, its computational
implementation and its incorporation implementation in an autonomous unmanned aerial vehicle
(UAV) This technology supports the operation of the LiDAR sensor using cameras and video and
image processing techniques to previously optimize the performance of subsequent algorithms that
use deep learning, reducing the number of computations without sacrificing much of the system
performance. On the other hand, by using computer vision techniques, the use of autonomous
vehicle hardware could be optimized, including techniques for processing scenarios with non-ideal
environmental conditions (rain, fog, darkness, and others), particularly for the Costa Rican reality.
The case studies to be applied of the Costa Rican reality in this project will be to map urban
infrastructure. The main contribution of this proposal is to integrate the algorithms developed from
this research for the generation of three-dimensional models from two-dimensional images in
conjunction with one or more autonomous unmanned aerial vehicles (UAV).This system will be
validated in the Costa Rican urban topography area, this area is chosen because the generation of
these models is complex due to the diversity of structures present in different environments which
do not always have ideal viewing conditions. Said validation will be carried out by comparing the
performance of the final product of this proposal against the collated digital elevation model,
generated by the project: Generation of flood patches of the upper basin of the Agua Caliente River,
carried out by the School of Agricultural Engineering in conjunction with the National Emergency
Commission (CNE).

Desarrollar un vehículo aéreo no tripulado basado en un
nuevo enfoque de procesamiento de imágenes, libre de un sensor LiDAR, para el
procesamiento confiable del entorno en condiciones de visión no ideales para la
topografía urbana costarricense.

Escuela de Matemáticas. Matemática aplicada: modelación, simulación, inteligencia artificial, análisis de
datos,visualización de información, optimización y aplicaciones a la ingeniería y a las ciencias. Área Académica
Ingeniería en Computadores: Sistemas embebidos, arquitectura y organización de computadores,inteligencia
artificial y robótica, redes y sistemas de comunicación para sistemas computacionales, interacción
humanomáquina,aplicaciones de la computación en diferentes dominios, computación de alto desempeño.