Notify vehicle driver about speed limits by analysing road signs in Sri Lanka.

Abstract

The number of motor traffic accidents are increasing day by day. One of the main reasons for accidents is over speeding. Over speeding also causes an uncomfortable journey and economic losses due to the traffic fines. There are speed limit signs to assist the driver to drive safe speed in the road. But most of the drivers will not drive under speed limits due to, forgetting the speed limit, unintentionally over speeding, careless driving or unnoticing the speed limit signs. Therefore, it’s better to inform driver about the speed limit as well as notify them when they are breaking the speed limit which leads to follow the speed limits in the road. Eventually passengers and driver will have a safe drive and driver will be able to avoid getting fines due to violating speed limits. Author proposed a mobile application to assist or guide drivers about the driving speed, which helps drivers to drive under suitable speed. Thus, passengers will have a comfortable and safe journey. The proposed mobile application uses the mobile phone’s camera and Global Positioning System (GPS). The application will get a frame or image from the smart phone’s camera and check that frame/image contain any circular shape objects by Circular Hougman Transformation (CHT) technique. Detected circles extracted and convert to a binary image and the Optical Character Recognition (OCR) done by using Tesseract engine, after removing the noise by using morphological operations such as Erosion and Dilation. The resultant text will use to verify whether that circular object is speed limit sign or not by checking that text contains similar text to “KMPH” text (“KMPH” is Part of a text appear in the speed limit sign). If the system identified any speed limit sign circles that circle’s center coordinates uses to extract the vehicle types board which uses to detect speed limit affecting vehicles. Extracted image converted to a binary image and uses Dilation to connect parts of a vehicle together. Then extract objects in that image after removing unwanted background. System will extract features of each object by using Histogram of Oriented Gradients (HOG) and feed into Error Correction Output Codes (ECOC) Model, which is trained using the same techniques and 219 sample speed limit sign images. Finally, if that affecting vehicle types board included the driver’s vehicle type then detected speed limit value stored in the system as new speed limit. The application use android.location package to detect vehicle speed. If measured speed is greater than the stored speed limit value then the system will inform that using text to speech library. Apart from that, system will announce new speed limit when the system detects a new speed limit sign which affected to the vehicle. The application is given us hopeful accuracy as well as performance. The whole process needs around 5.23 seconds and 73.25% accuracy. Since most of the current mobile phones have multiple CPU cores, detection of speed limit sign can be done by using parallel computing. Accordingly, it could able to improve performance while not decreasing the accuracy. Users could able get better results taking few actions to get clear images such as using a clear windscreen, keeping the camera in correct angle, etc.