AI DRIVER MONITORING SYSTEM

Project to monitor driver behiviour while driving. If the application detects driver rules violation done by driver then it sends a warning/image to a Telegram/Whats App/Email through REST API.

PROBLEM STATEMENT

School Buses: Driver drowsy, driver talking on cellphone while driving, driver/conductor smoking in the bus etc

Prerequisites

For Application:

opencv-python==4.4.0.44
requests==2.24.0
configparser==5.0.1
tensorflow==2.3.1
scikit-learn==0.23.2
wget==3.2
pygame==2.0.0
dlib==19.21.0
imutils==0.5.3

DETECTION RESULT

^{AI DRIVER MONITORING SYSTEM}

Vision

It has three vision based functionalities right now:

1. Detects if a Driver is drowsy.
2. Uses a custom trained model based on yolov4 Darknet architecture to detect if a person is smoking while driving.
3. Uses a custom trained model based on yolov4 Darknet architecture to detect if a person is using cell phone while driving.

When there is no detection of Driver Rules Violation

Face detection

Earlier, Dlib's frontal face HOG detector was used to find faces. However, it did not give very good results. In face_detection different face detection models are compared and OpenCV's DNN module provides best result and the results are present in It is implemented in AI_DRIVER_MONITORING.py and is used for tracking eyes, mouth opening detection and head pose estimation.

Facial Landmarks to detect if driver is Drowsy

To give landmarking on the eyelashes and if two lines on upper and lower eyelashes intersect the the application gives a warning message indicating the driver is Drowsey.

It is implemented in AI_DRIVER_MONITORING.py and is used for tracking eyes,and eyelashes.

Smoking and cellphone Detection while driving

AI_DRIVER_MONITORING.py is for detecting drivers using mobile phones and smoking while driving . It is based on a custom trained yolov4 architecture based Darknet Model.

Tips to train a custom based yolov4 based object Detection model you can refer to https://blog.roboflow.com/train-a-tensorflow2-object-detection-model/