In-vehicle Driving Quality Monitoring
Robotics and Intelligent Vehicle Automation Lab (RIVAL)
- Built by Dong Chen, Zhaojian Li from Michigan State University
- Started on Oct.19, 2019, Lastly updated on Sept.14, 2021
Overview
This project aims at building a on-device APP used to asist human drivers. This APP combines three basic functions: object detection(vehicle, traffic light, stop sign, pedestrain), lane deviation warning and distance estimation.
Motivation:
To be added...
Part1. Project Built Offline
Object Detection Module
We use the deep learning methods to do object detection. To be specific, we use the YOLO-v3 model to do object detection, here we are only curious about traffic-related objects, such as vehicles, pedestrain, traffic lights and stop signs.
Architecture of Yolo-v3 model
Lane Deviation Module
Considered limited computing resources on mobile devices (smart phones), we adapt the convential computer vision methods.
Modification logs:
- Delete the display code for "intermediate pipeline images".
- Simiplify codes.
- Problems with road curvature and offset values are always positive.
Distance Estimation Module
47o FOV len.
Distance Estimation
d = F_c * H_c / (y_b - y_h)
Demos:
For privacy issues, there are few open resources for dash camera videos. We will show our application by three different video demos.
-
Stop sign detection
-
Traffic light detection
-
Lane deviation detection
Reference
- YOLOv3: An Incremental Improvement
- What’s new in YOLO v3?
- Integration of Vehicle Detection and Distance Estimation using Stereo Vision for Real-Time AEB System
- Robust Range Estimation with a Monocular Camera for Vision-Based Forward Collision Warning System
- Advanced Lane Finding
- Lane Departure Warning System for Autonomous Driving
Part2. Project Built On Android
To be added...