The full text of this article/chapter:
Real time car detection in images based on an AdaBoost machine learning approach and a small training set
by Milos Stojmenovic
12th International Workshop on Systems, Signals and Image Processing (IWSSIP), Vol. 1, No. 1, 2005
Abstract: Our primary interest is to build fast and reliable object recognizers in images based on small training sets. This is important in cases where the training set needs to be built mostly manually, as in the case that we studied, the recognition of the Honda Accord 2004 from rear views. Our experiments indicated that the set of features used by Viola and others for face recognition was inefficient for our problem; therefore, each object requires its own custom-made set of features for real time and accurate recognition. We described a set of appropriate feature types for the considered car recognition problem, including a redness measure and dominant edge orientations. The existing edge orientation bin division was improved by shifting so that all horizontal (vertical, respectively) edges belong to the same bin. This feature set was a basis for building a fast and reliable car recognizer based on small training set, consisting of 155 positive and 760 negative images. It detects back views of Honda Accords with a 98.7% detection rate and 0.4% false positive rate on the training set, and with 89.1% detection rate and a 1.48 x 10-6 false positive rate on a test set of 106 images containing roughly 17.5 million tested sub windows.
is only available to individual subscribers or to users at subscribing institutions.
Pay per view: If you are not a Subscriber and you just want to read the full contents of this article, please click here to purchase online access to the full-text of this article. Please allow 3 days + mailing time. Current price for article is US$38.00
Members of the Editorial Board or subscribers of the 12th International Workshop on Systems, Signals and Image Processing (IWSSIP), that have been redirected here, please click here if you have IP-authentication access, or check if you have a registered username/password subscription with Inderscience. If that is the case, please Login: