Publicação
Automotive LiDAR based on active stereoscopy with extended range
| Resumo: | The pursuit to improve vehicle safety has led the automotive industry to develop various technologies, ranging from the simple seat belt to the Adaptive Cruise Control. While these can reduce the number of accidents, the main goal is to eliminate them altogether, based on a variety of sensors that give the vehicle the ability to perceive its surroundings. For this, the three main automotive vision sensors are the camera, RADAR and LiDAR. From all these sensors, LiDAR has the unique ability of producing a highly detailed point cloud, thus enabling the 3D object detection. Despite its relevance, this sensor is still one of the most expensive, which has hampered its adoption on the automotive market. In order to enable the adoption of LiDAR, this work proposes an approach based on active stereoscopy, capable of generating a point cloud similar to that of a LiDAR. This approach, called LiDART, has the advantage of using mass-produced components, thus making the sensor affordable. However, there are four main challenges in active stereoscopy: noise, correspondence, centroiding and calibration. This work focuses on noise, with the aim of developing a method that enables the sensor to have a longer range. As the distance between the active stereo cameras and the target increases, the dots reflected by the target become dimmer, since their energy decreases with the square of the distance. To prevent the spots from becoming unnoticeable at very short distances, we propose adding a second pattern projector identical to the original, but translated. The additional dot pattern doubles the number of projected dots. At close range, the individual points of each pair of dots can be resolved by the active stereo cameras, doubling the density of the point cloud. At a longer range, the dots merge into one, which results in the capture of twice the number of photons, thus minimizing noise. The main result of this work is the numerical and experimental validation of this operating principle, which results in an increase in the sensor’s range by a factor of √2. |
|---|---|
| Autores principais: | Rodrigues, Gonçalo Oliveira |
| Assunto: | Autonomous vehicles LiDAR Active stereoscopy Noise Laser Range |
| Ano: | 2024 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso embargado |
| Instituição associada: | Universidade de Aveiro |
| Idioma: | inglês |
| Origem: | RIA - Repositório Institucional da Universidade de Aveiro |
| Resumo: | The pursuit to improve vehicle safety has led the automotive industry to develop various technologies, ranging from the simple seat belt to the Adaptive Cruise Control. While these can reduce the number of accidents, the main goal is to eliminate them altogether, based on a variety of sensors that give the vehicle the ability to perceive its surroundings. For this, the three main automotive vision sensors are the camera, RADAR and LiDAR. From all these sensors, LiDAR has the unique ability of producing a highly detailed point cloud, thus enabling the 3D object detection. Despite its relevance, this sensor is still one of the most expensive, which has hampered its adoption on the automotive market. In order to enable the adoption of LiDAR, this work proposes an approach based on active stereoscopy, capable of generating a point cloud similar to that of a LiDAR. This approach, called LiDART, has the advantage of using mass-produced components, thus making the sensor affordable. However, there are four main challenges in active stereoscopy: noise, correspondence, centroiding and calibration. This work focuses on noise, with the aim of developing a method that enables the sensor to have a longer range. As the distance between the active stereo cameras and the target increases, the dots reflected by the target become dimmer, since their energy decreases with the square of the distance. To prevent the spots from becoming unnoticeable at very short distances, we propose adding a second pattern projector identical to the original, but translated. The additional dot pattern doubles the number of projected dots. At close range, the individual points of each pair of dots can be resolved by the active stereo cameras, doubling the density of the point cloud. At a longer range, the dots merge into one, which results in the capture of twice the number of photons, thus minimizing noise. The main result of this work is the numerical and experimental validation of this operating principle, which results in an increase in the sensor’s range by a factor of √2. |
|---|
Atividades financiadas
Não foram encontrados projetos financiados.