Improvement of the Efficiency of Object Detection

Many great object detection algorithms have been developed in the recent years, and the precision and accuracy have improved. However, the computation efficiency was not always the main emphasis in related research efforts.... [ view full abstract ]

Many great object detection algorithms have been developed in the recent years, and the precision and accuracy have improved. However, the computation efficiency was not always the main emphasis in related research efforts. Most state-of-the-art object detection algorithms rely on Graphics Processing Units (GPUs) to achieve real-time performance. In other words, good Frame per Second (FPS) performance. But in real-world applications, most hardware configurations do not have such high computational power, and some embedded systems are only equipped with a single core or dual-core processor. Furthermore, in some cases, the equipment has limited electric power, so they also need a lighter algorithm, such as a drone.

In this paper, we focus on improving the efficiency of an object detection algorithm based on hand-crafted features using only the Central Processing Unit (CPU). We compare different feature selection versus computational efficiency and accuracy. Finally, we hope to give other researchers some useful information while balancing between computational efficiency and accuracy.