In this final work, a Python script was developed for the automatic detection of the horizon in videos using a modified code template that uses basic image processing methods for detecting the horizon in each frame. The starting point was the function detect_horizon_line, which was implemented in the given template for detecting the horizon in individual images. As part of the work, the script video_horizon.py was created, enabling the processing of an entire video in real time. The modified function from the template was applied to each frame, and the result was displayed visually in the form of a red horizon line. In order to increase the accuracy and capabilities of the program, improvements were introduced such as line stabilization and the limitation of extreme values. Tests were conducted on videos with various characteristics, such as lighting level, contrast, and camera movement, after which the advantages and limitations of such approach were analyzed. Finally, possibilities for further development and improvement of the program were proposed, including the use of color, the application of machine learning as well as some others.

U ovom završnom radu razvijena je Python skripta za automatsku detekciju horizonta na videozapisima pomoću dorađenog predloška koda koji koristi osnovne metode obrade slike za detekciju horizonta na svakom kadru. Polazišna točka bila je funkcija detect_horizon_line, koja je bila implementirana u zadanom predlošku za detekciju horizonta na pojedinačnim slikama. U sklopu rada izrađena je skripta video_horizon.py, kojom je omogućena obrada cijelog videozapisa u stvarnom vremenu. Na svaki kadar primijenjena je dorađena funkcija iz predloška, a rezultat se prikazivao vizualno u obliku crvene linije horizonta. Radi povećanja točnosti i mogućnosti programa, uvedena su poboljšanja poput stabilizacije linije i ograničenja ekstremnih vrijednosti. Provedena su testiranja na videozapisima različitih karakteristika, kao što su razina osvjetljenja, kontrast i pomicanje kamere, te su analizirane prednosti i ograničenja takvog pristupa. Na kraju su predloženi mogućnosti daljnjeg razvoja i poboljšanja programa, uključujući korištenje boje, primjenu strojnog učenja i drugi.