
This master thesis describes the most famous databases that contain a record of the actual lines of the traffic lane in the frames of video signals recorded while driving. An overview of the existing methods for the detection of lane lines based on deep learning is given. Then, the performance analysis of the three lane line detection methods based on deep learning was made on three different datasets. The method of implementation, testing and evaluation of trained models on different datasets is described, and a critical review of the obtained results is given. The SCNN-VGG16 model performed best when trained and tested on the same dataset. The accuracy is similar to that of the SCNNResNnet34 and the SCNN-ERFNet models, but the smaller number of FP examples produced by the SCNN-VGG16 model causes a more accurate prediction of traffic lane lines. If the test set is changed during model testing, the results are different for each dataset. For the CULane dataset, the SCNNVGG-16 model proved to be the most accurate. For TuSimple and LLAMAS test sets, the SCNNERFNet proved to be the most adaptive to test set change.
U radu su opisane najpoznatije baze podataka koje sadrže zapis o stvarnim linijama vozne trake u okvirima video signala snimljenim u vožnji. Dan je pregled i postojećih metoda za detekciju linija vozne trake zasnovanih na dubokom učenju. Potom je napravljena analiza performansi triju metoda za detekciju linija vozne trake zasnovanih na dubokom učenju na trima različitim skupovima podataka. Opisan je način implementacije, testiranja i evaluacije istreniranih modela na različitim skupovima podataka te je dan kritički osvrt na dobivene rezultate. SCNN-VGG16 model se pokazao najboljim kada se trenira i testira na istom skupu podataka. Preciznost je slična kod SCNN-ResNnet34 i SCNN-ERFNet modela, no manji broj FP primjera kod SCNN-VGG16 modela uzrokuje točnije predviđanje linija vozne trake. Ukoliko se promjeni testni skup prilikom testiranja modela, rezultati su drugačiji za svaki podatkovni skup. Za CULane podatkovni skup SCNN-VGG-16 model se pokazao kao najprecizniji. Za TuSimple i LLAMAS testni skup, SCNN-ERFNet se pokazao kao najprilagodljiviji promjeni testnog skupa.
TEHNIČKE ZNANOSTI. Računarstvo. Umjetna inteligencija., SCNN, IoU, linije vozne trake, datasets, ERFNet, ResNet34, baze podataka, deep learning, VGG16, duboko učenje, TECHNICAL SCIENCES. Computing. Artificial Intelligence., lane lines
TEHNIČKE ZNANOSTI. Računarstvo. Umjetna inteligencija., SCNN, IoU, linije vozne trake, datasets, ERFNet, ResNet34, baze podataka, deep learning, VGG16, duboko učenje, TECHNICAL SCIENCES. Computing. Artificial Intelligence., lane lines
| selected citations These citations are derived from selected sources. This is an alternative to the "Influence" indicator, which also reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically). | 0 | |
| popularity This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network. | Average | |
| influence This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically). | Average | |
| impulse This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network. | Average |
