Bird’s Eye View – A Primer to the paradigm shift in Autonomous Robotics
June 30, 2023
Introduction:
In the world of autonomous vehicles, Perception is a critical component that plays a pivotal role in ensuring safe navigation, object detection and informed decision-making of the vehicles. There are many advanced and sophisticated perception systems available for autonomous vehicles and the one approach gaining significant traction is Bird’s Eye View (BEV) perception. BEV perception offers valuable insights for navigation, object detection, and path planning by providing a top-down perspective of the surroundings.
Bird’s Eye View:
Bird’s Eye View, also known as a top-down view or overhead view, is a representation of the environment from an elevated perspective. In the context of autonomy, BEV refers to a visual representation of the surrounding scene of a subject as seen from above, as if observed by a bird flying overhead.
The creation of a BEV usually involves processing sensor data, such as LIDAR, camera images or combination of multiple sensors, to generate a top-down representation aligned with the ground plane. By transforming the sensor data into a different coordinate system, BEV enables a comprehensive view that captures a wider field of view and rich geometric information about the scene.
Perception View (PV) to Bird’s Eye View (BEV):
Vision centric BEV perception remains a long-standing challenge for the cost-effective autonomous driving systems, as the cameras are typically mounted on ego-vehicles parallel to the ground and facing outwards. The images are captured in a Perspective View (2D viewpoint of an observer), which are mostly orthogonal to BEV and the transformation from PV to BEV is commonly referred to as Inverse Perspective Mapping (IPM).
With advancements in Sensor Technology & Machine Learning, there are several Convolutional Neural Network (CNN) based approaches available to convert the Perception View to Bird’s Eye View. The following is a real-world visualization of PV to BEV using a CNN:
What is Occupancy Grid Map & How does it relate to BEV?
Occupancy Grid Map (OGM) is another popular technique in the field of autonomous systems to represent and model the environment of a vehicle or a robot. It provides a structured grid-based representation of the environment by dividing it into a grid of cells. Each cell in the grid represents a specific region or location within the environment.
The OGMs are generated by integrating the data from various sensors such as LiDAR, Cameras, or RADAR. Based on the sensor data, the occupancy state of each grid cell is updated. Each cell is associated with a probability value indicating the likelihood of occupancy.
Both the Bird’s Eye View (BEV) and Occupancy Grids are representations of the environment that can be used in the field of autonomous systems. While the processes of generating BEV and OGM differ, they both utilize a grid structure to represent the environment. In both techniques, the environment is divided into cells within a grid, with each cell corresponding to a specific location within the environment. Both the representations are utilized extensively in the field of autonomous systems due to their significant applications.
Applications of BEV Perception in Autonomy:
- Autonomous Driving: BEV perception can be used to identify the tracks & lane markings on the road which helps the autonomous vehicles to understand the road and stay within the designated lanes. BEV enables the detection & tracking of objects around the vehicles, such as obstacles, pedestrians, and various other objects on the road. It also helps the autonomous vehicles in path planning, parking assistance & traffic flow analysis etc.
- Robotics: Robots equipped with BEV perception can capture a comprehensive view of the environment and analyse it for security monitoring, anomaly detection and environmental monitoring etc. It can also be utilized to coordinate multiple robots operating in the same area, enabling them to avoid collisions and collaborate on complex tasks such as drones performing a light display.
Conclusion:
The future of BEV perception holds promising prospects for enhancing the perception capabilities of autonomous systems, leading to safer and more efficient transportation & usage in a wide range of applications across various industries. The approaches to convert the Perception View to Bird’s Eye View & its relation to Segmented Occupancy Grid Map will be discussed in successive blogs.
