Speed limiting sign recognition system using histogram of oriented gradients method and K-nearest neighbor classification based on raspberry pi

Abstract

The dominant transportation vehicle in Indonesia is manual transportation. This type of transportation is controlled by the driver himself. Cases of traffic accidents are increasing due to lack of awareness of driving safety and security. The biggest factor in accidents is human error. One accident caused by human error such as a driver who lost speed control, because he ignored the maximum and minimum speed limiting signs. Therefore, the solution for this problem is creating a warning systems that can be used for recognizing maximum and minimum speed limiting signs. The system uses a raspberry pi camera to capture images then be detected and recognized the speed sign. If the system manages to recognize the signs according to the actual conditions traversed by the driver, it will get notification of speed sign figures in the form of sound from the speakers. The study applied the Histogram of Oriented Gradients (HOG) method to obtain the characteristic feature extraction from the sign, then classify it using the K-Nearest Neighbor (K-NN) method. Classification testing using K-NN consist of 650 training data and 48 test data that are comes from six sign types, there are Maks 20 km/h, Max 25 km/h, Max 30 km/h, Max 40 km/h, Max 50 km/h, Min 20 km/h). The average accuracy values is 97.91% for k=1 and 2. Meanwhile, accuracy of k = 3, 4 and 5 have similar value, that is 95.83%. The average time of computing the system to recognize objects 897 milliseconds. The average result of recognition based on the best k value is 97.91%.