Visual Navigation using Neuromorphic Camera
- Paper number
GLEX-2025,9,1,4,x93343
- Author
Mr. Mijaz Mukundan, Vikram Sarabhai Space Centre, ISRO, Thiruvananthapuram, India
- Coauthor
Mr. Rithin Mohan, Vikram Sarabhai Space Centre, ISRO, Thiruvananthapuram, India
- Year
2025
- Abstract
Visual navigation is a challenging problem in the field of computer vision which involves using visual data to compute motion parameters like translational and rotational velocities and structural information of the scene. Traditional cameras produce a high amount of data at a low frame rates. Thus, it becomes a computation intensive task to obtain these parameters for fast moving bodies, which is not desirable for a spacecraft. We propose a novel way to perform visual navigation using neuromorphic/event cameras for spacecraft landing scenarios. Neuromorphic cameras are visual sensors which only respond to local changes in brightness. Traditional cameras use a shutter to capture images frame by frame, and in each frame, data is captured in every pixel in its sensor array. Event cameras work differently. Here each pixel in the sensor array works independently and detects changes in intensity of light in that pixel alone. Our work demonstrates ego motion estimation of a landing spacecraft using event data. This method will allow for a high-performance sensing of motion, using the limited processing power, which can augment the spacecraft navigation capabilities. In this work, event data was generated from a simulated landing video. A block search algorithm was implemented to estimate motion fields from this simulated event data. The block size, shape and cost function for matching were chosen based on experimental results. This algorithm is FPGA compatible and can be parallelized for efficient one-shot computing of the motion field. Finally, these motion fields were used to estimate the unit translational velocity vector and rotational velocity vector using a least squares optimization approach. This method provides additional sensing capabilities to a spacecraft trying to land on an unknown terrain and is computationally efficient enough to be deployed onboard. We believe that this approach combined with other navigation sensors using sensor fusion can provide a robust solution to the problem of ego motion estimation of a spacecraft.
- Abstract document
- Manuscript document
GLEX-2025,9,1,4,x93343.pdf (🔒 authorized access only).
To get the manuscript, please contact IAF Secretariat.