Real - Time Image Stitching for Application in Advanced Driver Assistance Systems

Traffic accident is one of serious problems that faced by many countries. Advanced Driver Assistance System (ADAS) is a safety system to help driver monitor environment around the vehicle. Most common sensor used in ADAS is camera. Given that collisions can occur from multiple directions, relying on a single camera is insufficient for comprehensive detection. To address this issue, we propose an Advanced Driver Assistance System (ADAS) that utilizes three cameras, each providing distinct images captured from different viewpoints. Using Speeded-Up Robust Features (SURF), images from two pair camera left-center and right-center will be stitched first. Then, results from both process will be stitched again to create final stitched image containing image from all camera. The stitched result will be process processed in Instance Segmentation using Yolov8-seg. From experiment result, our system show good performance and promising to increase camera field of view with FPS of 10.37. Our segmentation model has mean Average Precision@50 (mAP50) 82%, Recall 76.5% and Precision 82.1%.