- Introducing a novel patent based on deep learning to improve defensive driving at night and in adverse weather conditions for Ford vehicles
- Convert day-time images to nighttime images
- Simulate different adverse weather models
Using You Only Learn One Representation (YOLOR) algorithm to determines the smallest possible bounding box where the taillights and brake lights of the vehicles are located
|
--- test
Before run models, unzip single-detector.rar
python train.py --batch-size 8 --img 640 640 --data coco.yaml --cfg cfg/yolor_p6.cfg --weights runs/train/yolor_p62/weights --device 0 --name yolor_p6 --hyp hyp.finetune.1280 --epochs 110
Test the detector:
python detect.py --source inference/example.mp4 --cfg cfg/yolor_p6.cfg --weights yolor_p6.pt --conf 0.25 --img-size 1280 --device 0
Change "source" value to "0" to use camera: i.e.
python detect.py --source 0
- Collecting Large-scale dataset from the rear view of vehicles along with labeling Region of Interest (ROIs) of taillights and brake lights
- Supporting two types of Dash-cam and Insta-360 Cameras
- Including four classes: a) running b) braking c) left turn d) right turn
- All model are selected from the products of Ford Motor Company