Abstract:
Traditional traffic scene prediction methods use current observations to predict the future states of visible objects (not in blind spots). Based on current observations of visible objects, this study predicted the future states of objects in blind spots (e.g., outside the field-of-view or occluded regions). We proposed a method that predicts blind spot vehicle appearances based on visible pedestrian behavior. We predict pedestrian behavior using a spatiotemporal 3D convolutional neural network. Pose estimation and semantic segmentation represent subtle pedestrian motions and their surroundings. We created two real-traffic video datasets for evaluation experiments. Cameras without ego-motions collect the datasets. We tested simpler and realistic traffic environments using the datasets. Experimental results suggest the following: (i) Our proposed method predicted vehicles from blind spots more than 1.5 s before they appeared, similar to humans. (ii) Ensembling explicit pose and semantic mask representations improved prediction performance. (iii) To predict driving car videos, fine-tune the models using ego-motion videos.
Note: Please discuss with our team before submitting this abstract to the college. This Abstract or Synopsis varies based on student project requirements.
Did you like this final year project?
To download this project Code with thesis report and project training... Click Here
