.png){kind=icon}
.png){kind=icon}
Return
A Continuous Camera Placement Optimization Model for Surround View
.png){kind=icon}
PRE
AI
DOI：10.1109/TIV.2023.3299199.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Increasing use of Advanced Driver Assistance Systems (ADAS) in autonomous vehicles is rising the demand for advanced perception systems. With more sensors being placed on the vehicle than ever, a need arises to optimize the placement of the sensors on the vehicle's body to maximize coverage at minimal cost. Camera Placement Optimization (CPO) methods tailored for multi-camera networks to maximize a vehicle's surrounding view coverage are limited. While existing CPO methods tend to sample the simulation space to work in the discrete domain using integer programming-based problem formulation, this article proposes a novel approach to optimize for camera poses for vehicle surround-view in the continuous domain using gradient-free blackbox optimization techniques by defining a non-linear objective function. Experimental results on more than 100 instances of real world 3D models of vehicles of various shapes and sizes show that the proposed method is effective in maximizing coverage for vehicle surround-view by a multiple camera network in a reasonable amount of time. Comparisons against a discrete CPO formulation show that the proposed method significantly improves coverage accuracy by optimizing poses of multiple cameras for vehicle surround-view.
Keywords:
Cameras
Optimization
Solid modeling
Three-dimensional displays
Sensors
Space vehicles
Visualization
Autonomous driving
black-box optimization
global optimization
optimal camera placement
surround-view vision
Journal
I
IF:
14.3
Papers:
1.2K
Citations:
1.2W