Closed fscbr closed 6 years ago
We should test two approaches. First an SVM classifier which is available already. I testes already real pictures and improved the classifer by finding a search area using bitmaps. More tests need to be done to understand the accuracy we gain
Secondly a FCN scene segmentation classifier For this task we have augmented train images available. The FCN classifier is in a first draft version available the vgg needed to be downloaded from 'https://s3-us-west-1.amazonaws.com/udacity-selfdrivingcar/vgg.zip'
the traffic light test video has to be splitted in train and test parts. the classifier has to be trained to have a high accuracy