Semantic Scene Change Detection: Evaluation through Classical & Machine Learning Algorithms
Publicerad
Författare
Typ
Examensarbete för masterexamen
Program
Modellbyggare
Tidskriftstitel
ISSN
Volymtitel
Utgivare
Sammanfattning
Scene change detection helps to detect changes in a pair of multitemporal images of
the same scene. We apply the concept of scene change detection to detect misplaced
objects in a passenger vehicle. Deep learning neural networks have been extensively
used in scene change detection. We study scene change detection using the classical
Watershed algorithm and machine learning algorithms. In machine learning, we
exploit the feature extraction capability of ResNet and Spatial Pyramid Pooling to
predict the scene change. The performance of the classical and machine learning
algorithms are also compared. The models are trained on a custom dataset and evaluated using the metrics, dice coefficient, mean intersection over union (mIoU) and
pixel accuracy. We infer that the machine learning model significantly outperforms
the classical model in terms of mIoU score.
Beskrivning
Ämne/nyckelord
scene change detection, machine learning, semantic segmentation, convolutional neural network, residual neural network, siamese network, spatial pyramid pooling