Real-time Image Processing on Low Cost Embedded Computers
In 2012 a federal mandate was imposed that required the FAA to integrate unmanned aerial systems (UAS) into the national airspace (NAS) by 2015 for civilian and commercial use. A significant driver for the increasing popularity of these systems is the rise in open hardware and open software solutions which allow hobbyists to build small UAS at low cost and without specialist equipment. This paper describes our work building, evaluating and improving performance of a vision-based system running on an embedded computer onboard such a small UAS. This system utilises open source software and open hardware to automatically land a multi-rotor UAS with high accuracy. Using parallel computing techniques, our final implementation runs at the maximum possible rate of 30 frames per second. This demonstrates a valid approach for implementing other real-time vision based systems onboard UAS using low power, small and economical embedded computers.
Summary
This master's thesis documents building, evaluating, and optimizing a vision-based autonomous landing system for small multirotor UAS using low-cost embedded computers and open-source hardware/software. It demonstrates how parallel computing and careful hardware–software integration enabled real-time image processing at the platform limit of 30 frames per second to achieve high-accuracy landings.
Key Takeaways
- Implement a vision-based autonomous landing pipeline on a low-cost embedded computer using open-source tools.
- Optimize image-processing workloads with parallelism and platform-specific acceleration to sustain 30 FPS.
- Integrate camera and onboard sensors on Embedded Linux to support reliable pose estimation and control.
- Evaluate system latency and landing accuracy with real-world flight tests and performance measurements.
- Apply hardware/software co-design principles to balance cost, weight, and compute for UAS payloads.
Who Should Read This
Embedded systems and robotics engineers or graduate students with some experience in embedded Linux and vision who want to build or optimize real-time vision systems for small UAS.
Still RelevantIntermediate
Related Documents
- Consistent Overhead Byte Stuffing TimelessIntermediate
- PID Without a PhD TimelessIntermediate
- Introduction to Embedded Systems - A Cyber-Physical Systems Approach Still RelevantIntermediate
- Design and Implementation of the lwIP Stack Still RelevantAdvanced
- Lessons in Electric Circuits - Volume III - Semiconductor TimelessIntermediate
