Improving Vehicle Perception Through Image Stitching: A Serial and Parallel Evaluation

Image stitching is important in a large number of applications, particularly enhancing decision making in vehicular systems by providing an expansive view from different perspectives. This work discusses the merit of serial versus parallel implementations of an image stitching algorithm which uses the ORB method for feature identification and the RANSAC algorithm for homography calculation. Serial implementation stitches procedures sequentially, while the parallel variant makes use of multicore processing to distribute tasks such as feature extraction, matching, and homography computation across multiple cores. We evaluate both methodologies using metrics such as runtime, with Python, OpenCV, and multiprocessing. The findings of this study reveal a notable acceleration in the parallel implementation, indicating its applicability for real-time functions within autonomous vehicles, where swift image processing is crucial. This comparative examination underscores the benefits of parallel computing in the realm of image stitching, aimed at improving vehicle perception and facilitating decision-making processes.