thesis: Recognition of Traffic Sign with Digital Image Processing and Machine Learning
This job has as objective to introduce a study of automatic traffic sign recognition system divided into five steps: aquisition, preprocessing with the use of HSV model and techniques such as limiarization, morphology and edge detection; segmentation with the use of generalized Hough transform; description with the use of Histograms of Oriented Gradients (HOG) and finally, classification with the use of Support Vectors Machine (SVM). The proposed algorithm is made with MATLAB platform.
The images for traning is in the below path
./Dataset/Training/*
The images are divided 11 classes of traffic sign:
- 30 - 30 km/h speed limit
- 50 - 50 km/h speed limit
- 60 - 60 km/h speed limit
- 70 - 70 km/h speed limit
- 80 - 80 km/h speed limit
- 100 - 100 km/h speed limit
- 120 - 120 km/h speed limit
- B - Do Not Enter
- C - No Passing (for any vehicle type)
- D - No Passing (by vehicles over 3,5 t)
- STOP
All images have been taken from GTSDB and they are only related to signs with circle format and red color. Each folder contains a spefic quantity of positive and negative examples. To run the training process execute de Training.m file in the root folder of the project.The traning process will generate the SVM model to each class of traffic sign and will put them in the below path.
./Results/TrainedSVMs/*
The test process is described in the image below.
The test process will use the scripts contained in the folders called HOG, Preprocessing, Segmentation and Templates. This process will generate the images of positive and negative segmented traffic sign in the path below:
./Results/Test
The recognition of all the segmented images with the svg will generate a .dat file for each SVM and the results folder. To run the test process, just execute de main.m file.