Traffic signs detection and recognition
How to detect and recognize the traffic signs using machine learning? In this work the Performance Analysis of Traffic Sign Detection and Recognition Techniques is given in detail. Automatic detection and recognition of road traffic signs is an essential task for regulating the traffic and guiding and warning drivers and pedestrians. Traffic signs have dual role: first, they regulate the traffic and second, indicate the state of road. They are installed at specific locations and appear with colors that contrast against road environment. These signs can be classified according to their color and shape and both these characteristics constitute their content.
Although people started working in this area in 1960s, but significant advances were made later in 1990s. For the recognition of road signs, few systems that have been developed so far are: Traffic Sign Recognition (TSR) , Automatic Target Recognition System , Driver Assistance System (DAS), Advanced Driver Assistance System (ADAS), and Driver Support Systems (DSS) are to name a few. The visibility of traffic signs is crucial for the drivers‟ safety. For example, serious accidents happen when drivers do not notice a stop sign.
In order to prevent these kinds of accidents, a group of traffic signs called warning signs are employed to warn road users, of such kinds of danger in the coming parts of road. Since human visual perception abilities depend on the individual’s physical and mental conditions, these abilities can be affected by many factors such as tiredness, and driving tension.
In bad weather conditions such as heavy rain showers, fog, drivers pay less attention to traffic signs and concentrate on driving. In night driving, visibility is affected by the headlights of traffic oncoming and drivers could easily be blinded. Also, red color is transformed into orange color because of lights. Hence, it is very important to have an automatic system which can detect and recognize these kinds of traffic signs. Giving this information to drivers in advance time can prevent accidents, save lives, and increase driving performance .
In Pakistan, the traffic states are different and are less studied from the intelligence perspective. Therefore, we are interested in studying and analyzing the performance of available techniques that will detect and recognize the traffic signs of Pakistan.
Literature review shows that even though a lot amount of research has been carried out on traffic sign detection based on color segmentation, but none of the published work has discussed about the sensitivity of different available color spaces for varying lighting and weather conditions individually. We have taken up this issue in this work. Also, we have tested how these color spaces affect detection phase when the distance of the traffic sign (target) varies from the camera mounted on the car.
For the recognition stage, first we have applied Histogram Oriented Gradient (HOG) for shape based feature extraction. Then the two classifiers: Support Vector Machine SVM and K-NN are used for reading the contents of the segmented traffic sign. The quality of output of these classifiers is computed through confusion matrix.
Traffic Signs in Pakistan
For doing In this work the Performance Analysis of Traffic Sign Detection and Recognition Techniques a proper research is done on Pakistani traffic signs. Color and shape, both provide useful information to drivers. Even when the contents are not easily readable, their color informs the drivers either the signs are informative or warning type. Unfortunately, the traffic sign databases are region or state specific.
There is no standard database available of traffic signs. Also, we can hardly find the work that presents the detailed analysis of color spaces effects on the detection stage and compares SVM and K-NN classifiers for recognition stage for Pakistani traffic signs. Table 1-1 lists different shapes and colors of Pakistan traffic signs. They are usually United Kingdom (UK) inspired because their outer rims are usually of red color.
Also read here