Automated Lane Changing ADAS:
Objective:
Automated Lane Changing in Advanced Driver Assistance System (ADAS) enhances vehicle safety and efficiency by automating the process of changing lanes. It addresses the need for reducing human error, improving traffic flow, and enhancing overall driving comfort. In this project, I developed a deep learning algorithm and optimized it for deployment of vehicle ECU.
Contributions:
- Trained a custom Fully Convolutional Network (FCN) for segmentation of free space and obstacle to indicate drivable area when changing lanes. The focus of the training was to improve the performance of the existing model near the boundary areas between obstacles and free space.
- Optimized the FCN using quantization techniques and created an inference pipeline in C++ for integration with ADAS software for automated lane changing.
Result:
- Improved FCN model performance by 12% in terms of IoU compared to the baseline model.
- Facilitated deployment of camera based automatic lane change feature for robust testing.