基于显著性检测的星载嵌入式跟踪系统

随着航天科技的不断发展，计算机视觉算法在卫星上的应用方兴未艾，为了实现更多的功能需求和应对可能的威胁，视觉目标跟踪作为其中基础但具有挑战性的任务更是至关重要。然而，目前已有的目标跟踪算法大多数算法只限于对图像序列进行跟踪。另一方面，受到硬件条件制约，很多优秀的算法因为复杂度较高很少被应用到星载嵌入式系统中。这些目标跟踪算法运行时，通常需要人为地给出目标的边界框。为了自动得到边界框，需要目标检测算法对输入图像进行运动目标检测。本文提出了一种基于显著性检测和相关滤波的单目标检测与跟踪一体化算法，并与嵌入式系统相结合，在搭载的TMS320C6678芯片上达到了2 048 pixel×2 048 pixel分辨率下24 fps的帧率。具体地，检测算法负责对图像进行预处理并获得边界框，然后目标跟踪算法给出目标在后续帧中的位置。为了验证算法在实际跟踪中的有效性，本研究搭建了一个由相机、DSP和云台组成的光学平台并进行了实验验证。在该系统中，DSP自动完成检测、跟踪、驱动云台和再检测任务，达到了很好的检测跟踪效果。

Satellite-borne embedded tracking system based on salient detection

With the development of aerospace science and technology,the application of computer vision algorithms on satellites is in the ascendant.In order to achieve more functional requirements and deal with possible threats,visual target tracking is a fundamental but challenging task.However,most existing algorithms only operate on pure image sequences.On the other hand,restricted by hardware conditions,few excellent algorithms are applied to satellite-borne embedded systems because of their high complexity.These pure object tracking algorithms usually need to artificially give the bounding box of the object.To get the bounding box automatically,an object detection algorithm is necessary.We propose an integrated detection and tracking algorithm for single target based on saliency detection and correlation filtering, which achieves 24 fps on DSP TMS320C6678 for the resolution of 2 048 pixel×2 048 pixel.The detection algorithm is responsible for preprocessing the image and obtaining the bounding box,then the object tracking algorithm gives the position of the object in the subsequent frames. In order to verify the effectiveness of our algorithm during the actual tracking,in this research,we build an optical platform consisting of a camera,DSP and Pan/Tilt/Zoom(PTZ) and test it.In this system,the DSP automatically completes the tasks of detection,tracking,driving the PTZ and re-detection,and achieves good results of detection and tracking.