DeepCar Team

Nighttime Driver Behavior Prediction

• Introducing a novel patent based on deep learning to improve defensive driving at night and in adverse weather conditions for Ford vehicles

Image-to-Image Translation

• Convert day-time images to nighttime images

• Simulate different adverse weather models

Single-stage Detector

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

Dataset

• 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