Low Latency Depth of Field Fusion System and Method Employing Fpga for Autonomous Driving

Autonomous driving technology imposes stringent requirements on the real-time performance and accuracy of environmental perception, and multi-focus image fusion is a key technology to enhance the performance of in-vehicle vision systems. In this paper, we propose a low-latency depth-of-field fusion system based on a field-programmable gate array (FPGA), which aims to solve the problems of low computational efficiency and high latency in traditional methods. The system adopts an improved ORB algorithm to achieve image alignment, enhances the clarity assessment capability for complex textures by introducing a four-direction gradient optimization Tenengrad operator, and utilizes a dynamic chunking strategy to achieve non-uniform spatial domain fusion. In terms of hardware design, the FPGA parallel pipeline architecture integrates feature matching, gradient computation, and fusion decision in a coprocessor, reducing the single-frame 1080P image processing delay to less than 5 ms. Experimental results show that the system improves the accuracy of clarity assessment by 15% in complex scenes such as low light, rain, and fog, effectively eliminates the “block effect,” and provides an efficient and robust visual perception solution for autonomous driving. This study provides a new technical foundation for the real-time application of multi-sensor data fusion in autonomous driving.