Computer Vision-Based Advanced Driver Assistance System (ADAS) Development
Achievements
- Real-time high implementation of computer vision-based ADAS system on Object Detection, Multi-Object Tracking, Lane Detection, Lane Departure Warning System, Lane departure warning system, Forward Collision Warning System.
Productization
- Developed ADAS functionalities, including FCWS, LDWS, and LKAS, using only visual sensors.
- Integrated object detection using YOLO models (YOLOv5 to YOLOv10, EfficientDet) and lane detection with Ultra-Fast-Lane-Detection-v2 in ONNX/TensorRT.
- Enhanced tracking capabilities with ByteTrack for multi-object tracking and trajectory prediction.
- Provided example scripts for lane and object detection using ONNX/TensorRT models for ease of deployment.
- Implemented model quantization techniques (e.g., float16) for optimized model size and performance.
- Developed a fully operational video inference pipeline with configurable lane and object detection models.
- Implemented a traditional image processing pipeline for lane detection using camera calibration, distortion correction, perspective transforms, and color-based thresholding techniques.
- Created a robust lane pixel detection algorithm that includes lane curvature calculation and vehicle position detection relative to lane center.
- Compared the performance of traditional image processing techniques with deep learning models for lane detection and tracking performance.
Publicity
