Automated Blender environment recreation and vanishing point detection for image stitching optimization

In today's digital era, automation is crucial for enhancing efficiency and customer satisfaction in various industries, notably transportation and logistics (T&L). This thesis investigates the effectiveness of image stitching within the context of T&L, focusing on improving quality and computational speed to address challenges such as reducing storage demand and bandwidth usage. Conducted during an internship at Datalogic in the R&D department in Pasadena, the study specifically examines a machine utilizing Phase correlation for image alignment. A simulated environment was constructed from images captured by this machine, facilitating evaluation under diverse configurations and conditions. Various aspects, including camera orientation and perspective distortion, were analyzed due to challenges posed by avoiding the use of computationally intensive algorithms, such as homography warping. Consequently, stitched images may exhibit perspective distortion due to phase correlation used for image alignment estimation. An innovative algorithm for vanishing-point determination was developed to mitigate the keystone effect, while enhancing image stitching by analyzing its relationship with phase correlation. Despite significant progress in improving image stitching for T&L machines, certain assumptions were made, such as neglecting roll and yaw angles of the camera, which warrant further investigation. Additionally, the study did not focus on determining confidence values for evaluating the validity of vanishing point estimation, suggesting avenues for future research. Overall, this thesis offers a potential solution to challenges associated with the Phase correlation algorithm in image stitching for a wide range of T&L machines, while also identifying areas for further exploration and refinement.