Publicação
Collision detection for UAVs using Event Cameras
| Resumo: | This dissertation explores the use of event cameras for collision detection in unmanned aerial vehicles (UAVs). Traditional cameras have been widely used in UAVs for obstacle avoidance and navigation, but they suffer from high latency and low dynamic range. Event cameras, on the other hand, capture only the changes in the scene and can operate at high speeds with low latency. The goal of this research is to investigate the potential of event cameras in UAVs collision detection, which is crucial for safe operation in complex and dynamic environments. The dissertation presents a review of the current state of the art in the field and evaluates a developed algorithm for event-based collision detection for UAVs. The performance of the algorithm was tested through practical experiments in which 9 sequences of events were recorded using an event camera, depicting different scenarios with stationary and moving objects as obstacles. Simultaneously, inertial measurement unit (IMU) data was collected to provide additional information about the UAV’s movement. The recorded data was then processed using the proposed event-based collision detection algorithm for UAVs, which consists of four components: ego-motion compensation, normalized mean timestamp, morphological operations, and clustering. Firstly, the ego-motion component compensates for the UAV’s motion by estimating its rotational movement using the IMU data. Next, the normalized mean timestamp component calculates the mean timestamp of each event and normalizes it, helping to reduce the noise in the event data and improving the accuracy of collision detection. The morphological operations component applies mathematical operations such as erosion and dilation to the event data to remove small noise and enhance the edges of objects. Finally, the last component uses a clustering method called DBSCAN to group the events, allowing for the detection of objects and estimation of their positions. This step provides the final output of the collision detection algorithm, which can be used for obstacle avoidance and navigation in UAVs. The algorithm was evaluated based on its accuracy, latency, and computational efficiency. The findings demonstrate that event-based collision detection has the potential to be an effective and efficient method for detecting collisions in UAVs, with high accuracy and low latency. These results suggest that event cameras could be beneficial for enhancing the safety and dependability of UAVs in challenging situations. Moreover, the datasets and algorithm developed in this research are made publicly available, facilitating the evaluation and enhancement of the algorithm for specific applications. This approach could encourage collaboration among researchers and enable further comparisons and investigations. |
|---|---|
| Autores principais: | Paulo, João Pedro Carvalho |
| Assunto: | Event-cameras UAVs Collision detection algorithm Ego-motion IMU Dynamic objects |
| Ano: | 2023 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade Nova de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório Institucional da UNL |
| Resumo: | This dissertation explores the use of event cameras for collision detection in unmanned aerial vehicles (UAVs). Traditional cameras have been widely used in UAVs for obstacle avoidance and navigation, but they suffer from high latency and low dynamic range. Event cameras, on the other hand, capture only the changes in the scene and can operate at high speeds with low latency. The goal of this research is to investigate the potential of event cameras in UAVs collision detection, which is crucial for safe operation in complex and dynamic environments. The dissertation presents a review of the current state of the art in the field and evaluates a developed algorithm for event-based collision detection for UAVs. The performance of the algorithm was tested through practical experiments in which 9 sequences of events were recorded using an event camera, depicting different scenarios with stationary and moving objects as obstacles. Simultaneously, inertial measurement unit (IMU) data was collected to provide additional information about the UAV’s movement. The recorded data was then processed using the proposed event-based collision detection algorithm for UAVs, which consists of four components: ego-motion compensation, normalized mean timestamp, morphological operations, and clustering. Firstly, the ego-motion component compensates for the UAV’s motion by estimating its rotational movement using the IMU data. Next, the normalized mean timestamp component calculates the mean timestamp of each event and normalizes it, helping to reduce the noise in the event data and improving the accuracy of collision detection. The morphological operations component applies mathematical operations such as erosion and dilation to the event data to remove small noise and enhance the edges of objects. Finally, the last component uses a clustering method called DBSCAN to group the events, allowing for the detection of objects and estimation of their positions. This step provides the final output of the collision detection algorithm, which can be used for obstacle avoidance and navigation in UAVs. The algorithm was evaluated based on its accuracy, latency, and computational efficiency. The findings demonstrate that event-based collision detection has the potential to be an effective and efficient method for detecting collisions in UAVs, with high accuracy and low latency. These results suggest that event cameras could be beneficial for enhancing the safety and dependability of UAVs in challenging situations. Moreover, the datasets and algorithm developed in this research are made publicly available, facilitating the evaluation and enhancement of the algorithm for specific applications. This approach could encourage collaboration among researchers and enable further comparisons and investigations. |
|---|