Abstract—In recent years, the object detection including the face detection becomes one of the most prosperous ﬁeldsin the computer vision. Along with it, highly precise detection algorithms and high-speed circuits with low electric power consumption are demanded by various camera systems. The most of mainstream object detection algorithms spend more time on learning general pattern of the objects in advance.However, there are many situations that cannot be detected by the prior leaning. In order to solve this problem, it is needed to prepare various detectors for special situations.But, for low-cost devices, it is not effective to detect all situations. In this paper, we propose an efﬁcient object detec-tion system adapting to input pattern of special scenes. Our method uses the kernel-based matching with low-resolution image to learn and detect object. This method optimizeskernels using Genetic algorithms for pproximating the distribution function more precisely, then detects objects by superposition of optimized kernels. However the algorithm requires long time to optimize the kernel. Therefore we estimate some parameters to improve the learning time anddetection rate by simulation. The simulation results show that these parameters are effective for getting the kernel to learn the image.