Implementation of unmanned aerial vehicle moving object detection algorithm on intel atom embedded system

Abstract

Unmanned Aerial Vehicles (UAV), which are commonly known as drones, are aircrafts that have no human pilot on board. A UAV always implements plenty sophisticated functions such as for military surveillance. Among all these complicated functions, one common task is normally implemented on a UAV which is the moving object detection algorithm. There are mainly two ways to implement this algorithm on a UAV, one is to use ground control station and another way uses on board processing method. Implementation of an on board processing unit on a UAV can eliminate the need of video streaming and can implement the computer vision algorithms on board. On board processing unit has mainly two options in implementing the moving object detection algorithm: hardware implementation and software implementation. Hardware implementation means the entire algorithm is designed and transformed into a hardware circuitry. Hardware implementation usually provides promising processing speed. However, whenever there is a slight change in the algorithm, there will be a huge redesign effort and development costs involved. Therefore, this research wants to prove that software implementation of the same algorithm by using a low power general purpose processor can achieve the same processing speed as the hardware implementation. Moreover, the processing speed of the algorithm can be further improved by applying appropriate code optimization techniques on the software program of the algorithm. Experimental results show that software implementation of the UAV moving object detection algorithm by using low power Intel Atom processor D2700 can achieve 30 frame per second processing speed. This research also proved that code optimizing techniques such as loop unrolling and Single Instruction Multiple Data (SIMD) can improve the processing speed of the algorithm up to 25 percent. In conclusion, software implementation of the UAV moving object detection algorithm, which requires low redesign effort and development cost, is capable of achieving the same processing speed provided by the hardware implementation.