Publication: Detección y geolocalización de objetos mediante visión artificial desde vehículos en movimiento
Loading...
Identifiers
Publication date
2021
Defense date
2021-07
Authors
Advisors
Tutors
Journal Title
Journal ISSN
Volume Title
Publisher
Impact
Export
Abstract
Las técnicas de la visión artificial han evolucionado y han ganado relevancia a lo largo de los
últimos años dentro del ámbito tecnológico, implementándose en una gran variedad de campos,
siendo uno de estos y el principal en el que será principalmente centrado este trabajo el de la
conducción autónoma.
En este proyecto se desarrollará un algoritmo capaz de realizar la detección y la geolocalización
de objetos captados desde diferentes cámaras situadas en un vehículo en movimiento, haciendo
uso de imágenes tomadas desde diferentes perspectivas, así como de diferentes metadatos que
proporcionen la información necesaria para ubicar a los objetos dentro del recorrido realizado
por el vehículo.
Para llevar a cabo estos objetivos, será necesario el análisis y filtrado de una base de datos que
proporcione toda la información necesaria para el uso del algoritmo, así como el estudio y
aplicación de técnicas de triangulación mediante visión artificial para ser utilizadas por el
algoritmo. Además de esto, se deberá diseñar e implementar el algoritmo que hará uso de los
datos filtrados y las técnicas de triangulación previamente estudiadas, teniendo el objetivo final
de medir la eficacia de su predicción y resultados con respecto a la posición real del objeto.
Como resultado final, se obtendrá un algoritmo que en base a una información concreta
introducida mediante ficheros en formato CSV, realizará el proceso de geolocalizar los objetos
contenidos en los datos obteniendo su posición dentro del recorrido realizado por el vehículo.
Además de estas coordenadas, el algoritmo realizará una representación gráfica con una
perspectiva de vista aérea del escenario y los objetos que lo componen, representando la
posición del vehículo en el momento en el que realizo cada una de las detecciones del objeto,
así como la posición predicha por el algoritmo del objeto y su posición real.
Como datos de salida adicionales, el algoritmo generará unos ficheros CSV con métricas que
midan la precisión de la predicción mediante el cálculo de errores comparando la posición real
del objeto y la predicha por el algoritmo.
Computer vision techniques have evolved and gained relevance over the last years within the technological field, being implemented in a great variety of fields, being one of these and the main one in which this work will be mainly focused on, that of autonomous driving. This project will develop an algorithm capable of detecting and geolocating objects captured from different cameras located in a moving vehicle, making use of images taken from different perspectives, as well as different metadata that provide the necessary information to locate the objects within the route taken by the vehicle. To achieve these objectives, it will be necessary to analyze and filter a database that provides all the necessary information for the use of the algorithm, as well as the study and application of triangulation techniques through artificial vision to be used by the algorithm. In addition to this, it will be necessary to design and implement the algorithm that will make use of the filtered data and the triangulation techniques previously studied, having the final objective of measuring the effectiveness of its prediction and results with respect to the real position of the object. As a final result, an algorithm will be obtained which, based on specific information introduced by means of CSV format files, will carry out the process of geolocating the objects contained in the data, obtaining their position within the route taken by the vehicle. In addition to these coordinates, the algorithm will perform a graphical representation with an aerial view perspective of the scenario and the objects that compose it, representing the position of the vehicle at the moment in which it performed each of the detections of the object, as well as the position predicted by the algorithm of the object and its actual position. As additional output data, the algorithm will generate CSV files with metrics that measure the accuracy of the prediction by calculating errors comparing the actual position of the object and the one predicted by the algorithm.
Computer vision techniques have evolved and gained relevance over the last years within the technological field, being implemented in a great variety of fields, being one of these and the main one in which this work will be mainly focused on, that of autonomous driving. This project will develop an algorithm capable of detecting and geolocating objects captured from different cameras located in a moving vehicle, making use of images taken from different perspectives, as well as different metadata that provide the necessary information to locate the objects within the route taken by the vehicle. To achieve these objectives, it will be necessary to analyze and filter a database that provides all the necessary information for the use of the algorithm, as well as the study and application of triangulation techniques through artificial vision to be used by the algorithm. In addition to this, it will be necessary to design and implement the algorithm that will make use of the filtered data and the triangulation techniques previously studied, having the final objective of measuring the effectiveness of its prediction and results with respect to the real position of the object. As a final result, an algorithm will be obtained which, based on specific information introduced by means of CSV format files, will carry out the process of geolocating the objects contained in the data, obtaining their position within the route taken by the vehicle. In addition to these coordinates, the algorithm will perform a graphical representation with an aerial view perspective of the scenario and the objects that compose it, representing the position of the vehicle at the moment in which it performed each of the detections of the object, as well as the position predicted by the algorithm of the object and its actual position. As additional output data, the algorithm will generate CSV files with metrics that measure the accuracy of the prediction by calculating errors comparing the actual position of the object and the one predicted by the algorithm.
Description
Keywords
Visión artificial, Geolocalización