Unmanned Air Vehicles (UAVs) enable the observation of hazardous areas without endangering a pilot. Observational capabilities are provided by on-board video cameras and images are relayed to remote operators for analysis. However, vibration and wind cause video camera mounts to move and can introduce unintended motion that makes video analysis more difficult. Video stabilization is a process that attempts to remove unwanted movement from a video input to provide a clearer picture.
This thesis presents an onboard video stabilization solution that removes high-frequency jitter, displays output at 20 frames per second (FPS), and runs on a Blackfin embedded processor. Any video stabilization algorithm will have to contend with the limited space, weight, and power available for embedded systems hardware on a UAV. This thesis demonstrates how architecture-specific optimizations improve algorithm performance on embedded systems and allow an algorithm that was designed with more powerful computing systems in mind to perform on a system that is limited in both size and resources. These optimizations reduce the total clock cycles per frame by 157 million to 30 million, which yields a frame rate increase from 3.2 to 20 FPS.
Unmanned Air Vehicles (UAVs) enable the observation of hazardous areas without endangering a pilot. Observational capabilities are provided by on-board video cameras and images are relayed to remote operators for analysis. However, vibration and wind cause video camera mounts to move and can introduce unintended motion that makes video analysis more difficult. Video stabilization is a process that attempts to remove unwanted movement from a video input to provide a clearer picture.
This thesis presents an onboard video stabilization solution that removes high-frequency jitter, displays output at 20 frames per second (FPS), and runs on a Blackfin embedded processor. Any video stabilization algorithm will have to contend with the limited space, weight, and power available for embedded systems hardware on a UAV. This thesis demonstrates how architecture-specific optimizations improve algorithm performance on embedded systems and allow an algorithm that was designed with more powerful computing systems in mind to perform on a system that is limited in both size and resources. These optimizations reduce the total clock cycles per frame by 157 million to 30 million, which yields a frame rate increase from 3.2 to 20 FPS.
Unmanned Air Vehicles (UAVs) enable the observation of hazardous areas without endangering a pilot. Observational capabilities are provided by on-board video cameras and images are relayed to remote operators for analysis. However, vibration and wind cause video camera mounts to move and can introduce unintended motion that makes video analysis more difficult. Video stabilization is a process that attempts to remove unwanted movement from a video input to provide a clearer picture.
This thesis presents an onboard video stabilization solution that removes high-frequency jitter, displays output at 20 frames per second (FPS), and runs on a Blackfin embedded processor. Any video stabilization algorithm will have to contend with the limited space, weight, and power available for embedded systems hardware on a UAV. This thesis demonstrates how architecture-specific optimizations improve algorithm performance on embedded systems and allow an algorithm that was designed with more powerful computing systems in mind to perform on a system that is limited in both size and resources. These optimizations reduce the total clock cycles per frame by 157 million to 30 million, which yields a frame rate increase from 3.2 to 20 FPS.