Recognition of Sign and Text Using LVQ and SVM

Traffic Sign Recognition (TSR) is used to regulate traffic signs, warn a driver, and command or prohibit certain actions. Fast real-time and robust automatic traffic sign detection and recognition can support and disburden the driver and significantly increase driving safety and comfort. Automatic recognition of traffic signs is also important for an automated intelligent driving vehicle or for driver assistance systems. Traffic signs or road signs are signs erected at the side of or above roads to give instructions or provide information to road users. There are many more other types of traffic signs such as special regulation signs, signs for direction and position, welcome sign etc. This report work aims to present a colour segmentation approach for traffic sign recognition based on Linear Vector Quantization (LVQ) neural network and also focuses on triangular edge detection and feature extraction based on Hough transformation and Histogram of Oriented Gradient (HOG) respectively. At first samples of images in different weather conditions are collected and then Red Green Blue (RGB) images are converted into Hue Saturation Value (HSV) colour space. The samples are then trained using LVQ depending on the hue and saturation values of each pixel and then tested for colour segmentation. The edges of the triangular segmented images are then detected using Hough Transformation. Then samples are taken to extract features using HOG. Finally they are trained and tested using Support Vector Machine (SVM) to get the output image. The algorithms were applied to around 100 sampled images which are taken in different Despite the varying conditions, the algorithms worked almost accurately in all situations and the success rate was quite satisfactory with a very good response time of a few milliseconds. Individual text characters are detected as Maximally Stable External Region (MSER) and are grouped into lines, before being interpreted using optimal character recognition (OCR). Recognition accuracy is vastly improved through the temporal fusion of text results across consecutive frames.