Quality evaluation of edge detection in a road image sequences

• Main Authors: Rodrigo B. de A. Gallis, Leonardo M. Pereir, Ricardo L. Barbosa, João F. C. da Silva
• Format: Article
• Language: English
• Published: Universidade Federal de Uberlândia 2004-12-01
• Series: Revista Brasileira de Cartografia
• Subjects: mobile mapping system; image sequences; edge detection; quality evaluation.
• Online Access: http://www.rbc.ufrj.br/_pdf_56_2004/56_2_02.pdf

Terrestrial mobile mapping systems map interest features along roads such as poles, traffic signs, curb lines, garbage cans etc. The lab work, concerned to the object reconstruction, consists of transforming the video into still images on which homologous points and features of the road sequence are selected and measured. By means of photogrammetric intersection the object coordinates of these features and points are computed for 3D reconstruction. Using Canny algorithm for the automatic edge detection in a road image sequence the article initially focuses on the empiric determination of the required parameters (standard deviation s and high Ta and low Tb threshold). Then it presents the quality in terms of displacement of the automatically detected edges similar to those visually (manually) selected straight features extracted by a human operator that takes them as correct, therefore, as reference for the automatic extraction comparison and the quality evaluation. The results of the tests are discussed and show that the quality of the automatic detection – measured by a quantity of rights and wrongs – vary accordingly to the empirically determined standard deviation and high and low thresholds and also to the image sequence environment (street or road).