PRSM

Persian Road Surface Marking (PRSM) Dataset: The dataset consists of over 6800 labeled images of Persian road surface markings in 18 popular classes. It also contains road surface markings under various daylight conditions such as sunny, sunset, and nighttime. Furthermore, this dataset contains images of marking signs in three different qualities, excellent, fair, and poor. The poor images have serious occlusion, motion blur, or are depreciated. You can download the official version of the PRSM dataset by clicking on the following link. http://display.sbu.ac.ir/submission_prsm-dataset/

Classes of captured road surface marking

Class name	Number of training samples	Number of testing samples	Total number of samples in each class	Proportion
Caution Text	3441	1474	4915	9.80 %
Caution Symbol	1277	547	1824	3.64 %
Yield line or Shark’s teeth	2463	1056	3519	7.01 %
Crosswalk	19525	8368	27893	19.93 %
Crosswalk Caution Text	163	70	233	0.46 %
Crosswalk Caution Symbol	826	354	1180	2.35 %
Forward	4023	1724	5747	11.45 %
Forward and Turn Left	674	289	963	1.92 %
Forward and Turn Right	1462	627	2089	4.16 %
School	760	325	1085	2.16 %
Slow	2737	1173	3910	7.79 %
Speed Bump	3758	1610	5368	10.70 %
Speed Limit	135	58	193	0.38 %
Stop	1241	532	1773	3.53 %
Stop Line	3931	1684	5615	11.19 %
Strain Speed	657	281	938	1.87 %
Turn Left	142	61	203	0.40 %
Turn Right	438	187	625	1.25 %
Total Number of Samples	47653	20420	68073	100 %

Annotation tool

To use annotation tools:

1. Place your images in the annotation tool's designated folder.
2. Run the application responsible for image tagging (e.g., tag_fig)

classification

Accuracy Comparison of Different Classifiers and Feature Extraction

Classifier	Accuracy	KNN	Length of Feature Vector
SVM	0.87	0.97	6400
HOG	0.85	0.98	9216
POEM	0.89	0.93	8496

Contact

Any questions or discussions are welcome! Please raise an issue (preferred), or send me an email.

Ali Asghar Sharifi [[email protected]]