Vehicle Detection on Embedded Single Board Computers

In this paper we present a vision-based vehicle detection pipeline for autonomous vehicles. This pipeline is based on Histograms of Oriented Gradients (HOG) and Support Vector Machines (SVM) for classification. Primary motivation to conduct this research is to evaluate the performance of vision intensive tasks on embedded single board computers which is used to process the real time video frames in vehicles. Complete system is implemented on two different embedded boards that is Raspberry Pi 3 and Odroid C2. We evaluated our proposed system using the KITTI Vision Benchmark Suite, a standard in computer vision research. The results show that the implemented pipeline reaches the accuracy up to 70% on Odroid C2 and 65% on Raspberry Pi 3, given the limited training dataset size of 1200 examples. Furthermore, these embedded systems don't have any on board GPU that may improve the system's performance.