-
Notifications
You must be signed in to change notification settings - Fork 15
Traffic Lights Detection and Classification
https://codebox.net/pages/image-augmentation-with-python
https://hci.iwr.uni-heidelberg.de/node/6132
https://github.com/bosch-ros-pkg/bstld
This dataset contains 13427 camera images at a resolution of 1280x720 pixels and contains about 24000 annotated traffic lights. The annotations include bounding boxes of traffic lights as well as the current state (active light) of each traffic light.
The camera images are provided as raw 12bit HDR images taken with a red-clear-clear-blue filter and as reconstructed 8-bit RGB color images. The RGB images are provided for debugging and can also be used for training. However, the RGB conversion process has some drawbacks. Some of the converted images may contain artifacts and the color distribution may seem unusual.
Training set:
- 5093 images
- Annotated about every 2 seconds
- 10756 annotated traffic lights
- Median traffic lights width: ~8.6 pixels
- 15 different labels
- 170 lights are partially occluded
Test set:
- 8334 consecutive images
- Annotated at about 15 fps
- 13486 annotated traffic lights
- Median traffic light width: 8.5 pixels
- 4 labels (red, yellow, green, off)
- 2088 lights are partially occluded
Captured (from the up-facing dash-cam in the simulator
https://www.dropbox.com/s/gp7vtdk8tjo65kc/TLdataset01.zip?dl=0
- 775 frames at the first two traffic lights.
- Zipped it is 252MB.
- Dimensions of each image: 800x600
- Format: PNG for high quality
- Notes: You'll find three folders from different sessions and within those are folders labeled red, yellow, and green. Also included is a reference frame to demonstrate what we see in the simulator versus what the dash-cam is seeing (looking up from a different perspective at the same moment). There is one partial frame from the test site to demonstrate the difference in image quality.
https://www.dropbox.com/s/87xark39qyer8df/TLdataset02.zip?dl=0
- 4499 images classified in 4 folders (red: 1733, yellow: 253, green: 645, unknown: 1868)
- Zipped it is 197MB.
- Format: PNG for high quality
- Dimensions of each image: 224x224
- Acquired through 10 simulator track circuits (8 forward, 2 counter)
- Includes long-range (>62m)
- Includes unknowns (no traffic light)
- Includes mis-predictions from training using the smaller dataset
Additional dataset with 423 yellow light samples (all within 75m): https://www.dropbox.com/s/7ld9h9vt9ctluib/TLdataset03.zip?dl=0
- Zipped 17MB
Additional dataset https://www.dropbox.com/s/dboj3lt4sa3ecj4/TLdataset04.zip?dl=0
- Zipped 26MB
- 622 images classified in 3 folders (red: 329, yellow: 36, green: 257)
- Includes mis-predictions from training on prior datasets
- the simulator only changes the state of the traffic light, which is coming up next (all the other traffic lights stay in status Red)
https://github.com/tensorflow/models/tree/master/research/object_detection
https://medium.freecodecamp.org/recognizing-traffic-lights-with-deep-learning-23dae23287cc
https://github.com/davidbrai/deep-learning-traffic-lights
https://static.googleusercontent.com/media/research.google.com/ru//pubs/archive/37259.pdf
https://carnd.slack.com/archives/C6NVDVAQ3/p1505631493000037
https://www.youtube.com/watch?v=ZPzHfzaCYDQ
https://www.isprs-ann-photogramm-remote-sens-spatial-inf-sci.net/III-1/191/2016/isprs-annals-III-1-191-2016.pdf