The following Project is Divided into the following Modules:
- Driver Side Detection
- Distraction Detection Using YOLOv4 Tiny
- Drowsiness Detection Using Histrogram of Orient Gradients, Facial Landmarks and Eye Aspect Ratio
- Road Side Detections
- Pedestrian Detection YOLOv4
- Traffic Sign Detection Using YOLOv4
The GPU Configurations for the model training were as follows:
The data of human hands, in the form of images, was acquired from the Open Images Dataset Version 6, by Google Open Source.
The performance metrics of the trained model were as follows:
The implementation workflow was as follows:
The Eye Aspect Ratio (EAR) was calculated as follows:
res.mp4
The dataset chosen for this was the PENN FUDAN Pedestrian Dataset. Pedestrians' data containing around 400 images with an average of 6 pedestrians per image was downloaded (approx. 2400 annotations). Several images (around 200) were also sourced from Google Images.
The trained model had the following metrics:
Indian traffic-sign datasets are limited. Available ones are private with restricted access. For this task, the GTSRB (German Traffic Sign Detection Benchmark) dataset containing 800 distinct images was used. Traffic signs were classified into 4 categories: Prohibitory, Danger, Mandatory and Other
