Indoor UAV Object Detection Algorithms On Three Processors: Implementation Test And Comparison

Nowadays Unmanned aerial vehicles (UAV) are used in various applications. One of the domains that they are recently hired on is remote sensing applications which they should detect and identify the object. Till today various algorithms are proposed but among them SIFT, SURF, ORB and BRISK are more famous but based on the author's resources their performance over the different processors with the same target image is not reported before which is aimed at this research paper. The mentioned algorithms are implemented on three types of hardware, Raspberry Pi3, Odroid C2, and Intel NUC, which were selected based on their availability and different applications in the robotic system. In order to compare the images of a common object, the common UAV structure was designed that can carry all the processors to different heights. Python programming language and OpenCV machine vision library have been used to implement each algorithm. After going through the described calibration steps, the performance of each algorithm in terms of accuracy and processing time on each processor with three cases of rotation modes, the scale and deviation are compared and analyzed based on the image obtained from the drone camera. The results obtained in this research show that the Intel NUC processor has a good processing time in all four algorithms, and the maximum accuracy for all cases of rotation, scale and skew belongs to SIFT. In addition to the mentioned comparison, due to the importance of the weight of accessories, a parameter called TW is defined to show the relationship between processing time and processor mass and the results are reported. The obtained results show that the algorithm's performance result in Raspberry Pi3 and Odroid C2 is very close and inferior to the Intel Nuc processor. But TW for Odroid C2 has even better results than Intel Nuc, which can interest UAV designers and professionals who need to use lightweight drones in remote sensing applications.