基于嵌入式GPU的红外弱小目标检测算法

Infrared weak small target detection algorithm based on embedded GPU

摘要: 红外弱小目标的目标像素少，目标对比度低，成像帧率高，图像数据量大，检测实时性强。针对红外弱小目标检测算法适合于GPU并行计算的特点，对其在嵌入式GPU平台Jetson TX2上进行了并行优化实现。在检测算法设计、内存访问、调试优化3个方面进行了优化设计。实验结果表明，对640×480像素分辨率的红外视频，并行优化后的目标检测算法能够在10 ms内完成计算，满足实时处理需求。

Abstract: The infrared weak small targets have few target pixels, low target contrast, high imaging frame rate, large amount of image data, and strong real-time detection. Aiming at the characteristic that the infrared weak small targets detection algorithm was suitable for the GPU parallel computing, the parallel optimization was implemented on the Jetson TX2 of the embedded GPU platform, and the optimized design was mainly reflected in the following three aspects: detection algorithm design, memory access, and debugging optimization. The experimental results show that for the infrared videos with a resolution of 640×480 pixels, the target detection algorithm after parallel optimization can complete the calculation in 10 ms, which meets the requirements of real-time processing.