Color Decision System Based on Deep Learning and Fuzzy Inference System.

Autonomous driving system has become a hot issue. Similar to autonomous driving system, Autonomous Guided Vehicle (AGV) works in industry field. It is controlled by its autonomous driving system. This system detects the driving environment by using sensors such as vision sensor, laser sensor, ultrasonic sensor and so on. Among these sensors, vision sensor can obtain various kinds of the information of the environment. In an image, which is captured by vision sensor, the environment is represented by its color, shape and so on. Vision-AGV where vision sensor is mounted can be controlled by the colorful sign like the traffic light. To recognize the sign for controlling the AGV, it is necessary to detect the sign and analyze the meaning of the this sign. Faster Regions with Convolutional Neural Network features (Faster R-CNN) is applied to detect the sign. To analyze the meaning of the color sign, RGB color space, which is well known and consists of primary colors, can be used. However, it is difficult to analyze the color sign by using only RGB color space. Therefore, the proposed method is designed for detecting and recognizing the color of the sign with CIE L*a*b* color space. To improve the effect of the analysis, RGB and CIE L*a*b* color space are combined. Based on RGB, CIE L*a*b* and Peak Signal-toNoise Ratio (PSNR), the color of the region which is extracted by Faster R-CNN is determined by the fuzzy inference system. As experimental results, the proposed method can detect the sign and analyze the color of this sign.