Surround view image generation network for heavy trucks

With the rapid advancement of the automotive industry and intelligent technologies, the growing significance of surround view systems (SVSs) in enhancing driving safety has become increasingly pronounced. However, when traditional SVSs are applied to heavy trucks with detachable containers, a series of challenges is encountered. First, the calibration process of conventional SVSs necessitates a specialized calibration site and involves intricate procedures, resulting in elevated human labor costs. Second, the large size of the vehicle body leads to insufficient camera field coverage, thereby impacting the quality of the surround view images. In addressing these challenges, we attempt to depart from conventional methodologies by introducing an innovative surround view image generation network that directly generates high-quality fisheye images through a learning-based approach. Within this method, a mapping table regression module is specifically designed to learn the coordinate mapping relationship between fisheye images and surround view images. A pixel position adjustment module is constructed to refine the mapping relationship, further enhancing image quality. In addition, by leveraging a pixel brightness adjustment module, noise points in the generated surround view images are successfully eliminated, thereby improving the overall image clarity. Experimental results validate that our approach not only obviates the need for specialized calibration sites but also produces images of higher quality compared with traditional methods. This innovative approach not only facilitates the generation of high-quality surround view images for heavy trucks with detachable containers but also presents a novel pipeline for the implementation of SVS.