Authors
Advisor(s)
Abstract(s)
Esta dissertação aborda o problema de detecção e desvio de obstáculos
"SAA- Sense And Avoid" em movimento para veículos aéreos. Em particular
apresenta contribuições tendo em vista a obtenção de soluções para
permitir a utilização de aeronaves não tripuladas em espaço aéreo não segregado
e para aplicações civis.
Estas contribuições caracterizam-se por: uma análise do problema de
SAA em \UAV's - Unmmaned Aerial Vehicles\ civis; a definição do conceito
e metodologia para o projecto deste tipo de sistemas; uma proposta de \ben-
chmarking\ para o sistema SAA caracterizando um conjunto de "datasets\
adequados para a validação de métodos de detecção; respectiva validação
experimental do processo e obtenção de "datasets"; a análise do estado da
arte para a detecção de \Dim point features\ ; o projecto de uma arquitectura
para uma solução de SAA incorporando a integração de compensação
de \ego motion" e respectiva validação para um "dataset" recolhido.
Tendo em vista a análise comparativa de diferentes métodos bem como
a validação de soluções foi proposta a recolha de um conjunto de \datasets"
de informação sensorial e de navegação. Para os mesmos foram definidos
um conjunto de experiências e cenários experimentais.
Foi projectado e implementado um setup experimental para a recolha
dos \datasets" e realizadas experiências de recolha recorrendo a aeronaves
tripuladas. O setup desenvolvido incorpora um sistema inercial de alta
precisão, duas câmaras digitais sincronizadas (possibilitando análise de informa
formação stereo) e um receptor GPS. As aeronaves alvo transportam um
receptor GPS com logger incorporado permitindo a correlação espacial dos
resultados de detecção.
Com este sistema foram recolhidos dados referentes a cenários de aproximação com diferentes trajectórias e condições ambientais bem como incorporando
movimento do dispositivo detector.
O método proposto foi validado para os datasets recolhidos tendo-se
verificado, numa análise preliminar, a detecção do obstáculo (avião ultraleve)
em todas as frames para uma distância inferior a 3 km com taxas de
sucesso na ordem dos 95% para distâncias entre os 3 e os 4 km.
Os resultados apresentados permitem validar a arquitectura proposta
para a solução do problema de SAA em veículos aéreos autónomos e abrem
perspectivas muito promissoras para desenvolvimento futuro com forte impacto
técnico-científico bem como sócio-economico. A incorporação de informa
formação de \ego motion" permite fornecer um forte incremento em termos
de desempenho.
This thesis addresses the problem of perception for detection and avoidance (\SAA - Sense And Avoid") of obstacles during the movement of unmanned aircraft. It presents speci c contributions to the search of solutions that allow the use of unmanned aerial vehicle (UAV - Unmmaned Aerial Vehicle) in a non-segregated air space for non-military purposes. These contributions are characterized by: the analysis of the SAA problem in civilian UAV's ; the de nition of the concept and corresponding methodology for the project of such systems; a benchmarking proposal for the SAA system based upon a group of adequate datasets for the validation of detection methods; experimental validation of the process itself along with the collection of datasets; the analysis of the state of the art in \Dim point features" detection; the design of an architecture for an SAA solution that incorporates the integration of ego motion compensation as well as the corresponding validation for a collected dataset. Since it is intended to compare di erent methods as well as validating their respective solutions, the gathering of sensorial and navigation information datasets has been proposed. A set of experiences and experimental scenarios have been established. To obtain the datasets, an experimental setup was designed and implemented along with data gathering actions using a conventional airplane. The developed setup includes a highly accurate inertial navigation system, two synchronized digital cameras (allowing the analysis of stereo information) and a GPS reception device. The target airplanes carry GPS devices with data logger that will provide the necessary spatial correlation of the detection results. With this system it was possible to collect data from approximation scenarios with di erent trajectories and environmental conditions along with movement of the detection device. The presented method was validated for the collected datasets and a preliminary analysis showed that the obstacle detection (glider aircraft) in every frame was achieved for a distance lesser than 3 Km, with success rates of 95% for distances raging from 3 to 4 Km. The presented results con rm the proposed architecture for the solution of the SAA problem in UAV's and opens a new set of promising perspectives in future developments with an important contribution in scienti c and technological as well as social and economical contexts. The inclusion of ego motion information resulted in increased performance of the method.
This thesis addresses the problem of perception for detection and avoidance (\SAA - Sense And Avoid") of obstacles during the movement of unmanned aircraft. It presents speci c contributions to the search of solutions that allow the use of unmanned aerial vehicle (UAV - Unmmaned Aerial Vehicle) in a non-segregated air space for non-military purposes. These contributions are characterized by: the analysis of the SAA problem in civilian UAV's ; the de nition of the concept and corresponding methodology for the project of such systems; a benchmarking proposal for the SAA system based upon a group of adequate datasets for the validation of detection methods; experimental validation of the process itself along with the collection of datasets; the analysis of the state of the art in \Dim point features" detection; the design of an architecture for an SAA solution that incorporates the integration of ego motion compensation as well as the corresponding validation for a collected dataset. Since it is intended to compare di erent methods as well as validating their respective solutions, the gathering of sensorial and navigation information datasets has been proposed. A set of experiences and experimental scenarios have been established. To obtain the datasets, an experimental setup was designed and implemented along with data gathering actions using a conventional airplane. The developed setup includes a highly accurate inertial navigation system, two synchronized digital cameras (allowing the analysis of stereo information) and a GPS reception device. The target airplanes carry GPS devices with data logger that will provide the necessary spatial correlation of the detection results. With this system it was possible to collect data from approximation scenarios with di erent trajectories and environmental conditions along with movement of the detection device. The presented method was validated for the collected datasets and a preliminary analysis showed that the obstacle detection (glider aircraft) in every frame was achieved for a distance lesser than 3 Km, with success rates of 95% for distances raging from 3 to 4 Km. The presented results con rm the proposed architecture for the solution of the SAA problem in UAV's and opens a new set of promising perspectives in future developments with an important contribution in scienti c and technological as well as social and economical contexts. The inclusion of ego motion information resulted in increased performance of the method.
Description
Keywords
Sistemas autónomos Veículos autónomos aéreos Percepção "Sense & Avoid" Visão Autonomous systems Unmmaned aerial vehicles Perception Vision
Citation
Publisher
Instituto Politécnico do Porto. Instituto Superior de Engenharia do Porto