Traffic-signage detection and recognition on K-means clustering and Support Vector Machine classification

Abstract

Traffic signage detection and recognition are essential components in the intelligent vision based transportation system by which in-vehicle is able to recognize the traffic signage in various shapes and interpret the contents such as speed limit, stop, and yield. The main challenge of traffic signage detection and recognition is to find a way to extract specific information from an image and classify it to the right category. Since the traffic signage images may contain shading and noise, the quality of the images can be varied from one to another. Besides, machine learning kernel for signage recognition is compute-intensive and requires extremely long precessing time to classify an image. In this project, an intelligent traffic signage detection and recognition system which uses k-means as signage detection and support vector machine (SVM) as classification model is proposed. The objective of this project is to develop a color segmentation algorithm which is invariant to different illumination levels and to optimized the performance of recognition algorithm in term of speedup. The new model has successfully implemented in the software environment to compared its performance with existing works. Experimental results show an overall detection rate of 97.6% is achieved. The linear kernel outperforms the RBF kernel by 3% with 92% of correct classification rate. Combining all the modules, the single frame processing time is successfully speedup by 333x and the memory utilization is reduced by 99.8%. The proposed method has excelled the previous works and is preferable for future hardware implementation.