Javascript must be enabled for the correct page display

Using Alpha trees for traffic sign recognition

Elhorst, Siebert (2020) Using Alpha trees for traffic sign recognition. Master's Internship Report, Computing Science.

[img]
Preview
Text
Using_Alpha_trees_for_traffic_sign_recognition.pdf

Download (12MB) | Preview
[img] Text
toestemming.pdf
Restricted to Registered users only

Download (96kB)

Abstract

In this research the α-tree is tested on recognition of traffic signs while maintaining decent speed. α-trees clusters pixels based on the difference in color of the selected tree depth, resulting in a tree of segmentation's. By grouping clusters based on alpha level, eventually a node with traffic sign might be found. When looking at the results, the speed of the given implementation is O(n log n) which is expected from the alpha tree. However the program could be further optimized such that it does not require more time of calculation than a car actually passing the traffic sign. 1024*768 images can be calculated within a few seconds, however for bigger images the implementation is not fast enough when looking for traffic signs. When looking at the recognition it is clear that leakage is a big reoccuring problem. Using the decision tree only a correct segmentation rate of 15% is achieved due to this leakage problem. However when using a Gabor filter this correct segmentation rate is drastically increased to 55%. Regardless of the recognition techniques it can be concluded that using a Gabor filter will increase the correct segmentation of an α-tree. The α-tree has the potential to recognize traffic signs but requires more research into preventing leakage (besides Gabor filtering), better efficiency such as parallelization and preventing unfortunate coloring.

Item Type: Thesis (Master's Internship Report)
Supervisor name: Wilkinson, M.H.F. and Zhang, X.
Degree programme: Computing Science
Thesis type: Master's Internship Report
Language: English
Date Deposited: 24 Jun 2020 12:41
Last Modified: 24 Jun 2020 12:41
URI: https://fse.studenttheses.ub.rug.nl/id/eprint/22232

Actions (login required)

View Item View Item