Real-Time Speed-Limit Sign Detection and Recognition Using Spatial Pyramid Feature and Boosted Random Forest

Abstract

Traffic-sign detection and recognition using computer vision is essential for safe driving when using an advanced driver assistance system (ADAS). Among the few types of traffic signs used, in this paper, we focus on the detection and recognition of speed-limit signs because such signs can ensure the safety of drivers and other road users, and facilitate an efficient traffic flow. To detect a speed-limit sign, we first choose the candidate regions for a speed-limit sign using the border color and apply sliding windows to the candidate regions using a two-class boosted random forest (BoostRF) classifier instead of simple random forest. To reduce the computational cost for the image pyramid, the optimal levels of scaling using the search area is adapted. Detected speed-limit signs are fed into the speed-limit sign classifiers based on the multiclass BoostRF. As the feature of the BoostRF, we use spatial pyramid pooling (SPP) based on oriented center symmetric-local binary patterns (OCS-LBP) because SPP is simple and computationally efficient, and maintains the spatial and local information by pooling the local spatial bins. The proposed algorithm was successfully applied to the German Traffic Sign Detection Benchmark (GTSDB) and German Traffic Sign Recognition Benchmark (GTSRB) datasets, and the results show that detection and recognition capabilities of the proposed method are similar or better than those of other methods.