Abstract :
In the field of object recognition, the most fundamental and important step is object detection. However, object detection in images requires a great deal of resources, for example, memory and computing power. If an image is enlarged, the computational complexity required for object detection becomes a large obstacle. An-integral-image-based method guarantees fast object detection. Once an integral image is generated, the speed of the object detection procedure remains fixed, regardless of the pattern region. Still, there are many recalculations during object detection on an integral image, and this constraint becomes an even greater issue if the image is enlarged. In this paper, we propose a customized-object detection method, which reuses pre-computed values and thereby reduces the number of computations required for object detection in comparison to methods using an integral image. In experiments comparing the two methods, the proposed method requires only 60% of the number of computations and detects objects 50% faster.
