Automatic road sign identification system with robustness to partial occlusion

Abstract

In recent years, automatic road sign identification system has attracted numerous research works with the possibility of using in autonomous or driver assistance system (ADAS). Research in road sign identification with occlusion, however is still lacking. Many existing techniques up to now that have been developed algorithms with the existence of occlusions produce inaccuracy that needs to be improved. Even though the occurrences of road signs with presence of occlusion are small, yet it is problem that needs to be addressed. An intelligent system for road sign identification that incorporated several different algorithms is proposed in this research to solve the problems. The algorithms consist of proposed HSV and RGB colors in detection part and ANN and PCA techniques in recognition part. The proposed algorithms are able to detect the three colored images of road sign namely Red, Yellow and Blue. These algorithms are then compared with each other to evaluate their performance. The hypothesis of this research is that road sign images can be used to detect and identify signs involved existence of occlusions and rotational changes. Each sign features are extracted using global feature extraction technique whereby the vertical and dimension size of sign are fixed to a standard size. These input features are used to be applied into neural network according to feed forward neural network technique using backpropagation training function. The sign image can be easily identified by the PCA method as it has been used in many application areas. Based on the experimental result, it shows that the HSV is robust in road sign detection with minimum of 88% and 77% successful rate for non-partial and partial occlusions images rather. For successful recognition rates using ANN can be achieved starts from 75-92% whereas PCA is in the range of 94-98%. The combination of HSV color-based detection and PCA generated faster processing time of 2.1s per frame for the overall identification process. The occurrences of all classes are recognized successfully is between 5% and 10% level of occlusions using PCA, whereas only 5% level of occlusions successful recognized using ANN.