Abstract
In the development of multi-vehicle cooperative hardware-in-the-loop (HIL) simulation platforms based on machine vision, accurate vehicle pose estimation is crucial for achieving efficient cooperative control. However, monocular vision systems inevitably suffer from limited fields of view and insufficient image resolution during target detection, making it difficult to meet the requirements of large-scale, multi-target real-time perception. To address these challenges, this paper proposes an engineering-oriented multi-camera cooperative vision detection method, designed to maximize processing efficiency and real-time performance while maintaining detection accuracy. The proposed approach first projects the imaging results from multiple cameras onto a unified physical plane. By precomputing and caching the image stitching parameters, the method enables fast and parallelized image mosaicking. Experimental results demonstrate that, under typical vehicle speeds and driving angles, the stitched images achieve a 93.41% identification code recognition rate and a 99.08% recognition accuracy. Moreover, with high-resolution image (1440 × 960) inputs, the system can stably output 30 frames per second of stitched image streams, fully satisfying the dual requirements of detection precision and real-time processing for engineering applications.