应用于纳型无人机视觉场景数据集构建的图像采集系统

2023,31(2):230-236
凤雷, 付洪硕, 吴瑞东, 陈浩林, 刘冰
哈尔滨工业大学
摘要:纳型无人机具有体积小、功耗低的优势,在物联网、智能感知等领域中有重要的应用前景。对于纳型无人机的机载目标检测和目标跟踪等任务,视觉场景数据集对于机载算法的训练具有重要作用。由于受到体积和功耗等因素的限制,使用无人机机载的视觉系统进行场景数据的采集存在传输帧率慢、传输中需要进行数据压缩等问题,进而导致构建的数据集与机载算法处理的实际图像不匹配。因此,为了构建一个高质量的数据集,该图像采集系统以匹配纳型无人机视觉场景数据为目的进行设计,系统通过低功耗图像传感器HM01B0制作的模组获取图像数据,通过ZYNQ的PL单元完成图像数据的处理和图像传感器的控制,PS单元完成图像数据存储和各任务间的调度。实验结果表明,该图像采集系统最高可以达到320*320分辨率下45帧/s的采集速率,与纳型无人机机载视觉系统图像获取速率相匹配,同时采集的图像质量清晰,能够满足纳型无人机视觉场景数据集构建的需求。
关键词:无人机;图像采集系统;ZYNQ;视觉场景数据集;FreeRTOS

Acquisition System Applied to the Construction of Visual Scene Data Set of Nano UAV

Abstract:The nano-UAV has the advantages of small size and low power consumption, and it has important application prospects in the fields of Internet of Things, intelligent perception and so on. Visual scene datasets play an important role in training airborne algorithms for such tasks as target detection and target tracking of nano-UAV. Due to the limitations of volume and power consumption, scene data collection using the UAV on-board visual system has some problems, such as slow frame rate, data compression required in the transmission, etc., which results in the data set built does not match the actual image processed by the airborne algorithm. Therefore, in order to build a high-quality data set, the image acquisition system is designed to match the visual scene data of nano-UAV . The system obtains image data through the module made by low-power image sensor HM01B0, completes image data processing and image sensor control through the PL unit of ZYNQ, and completes image data storage and scheduling between tasks by the PS unit. The experimental results show that the image acquisition system can reach a maximum acquisition rate of 45 frames/s at 320*320 resolution, which matches the acquisition rate of the image of the nano-UAV on-board visual system, and has a clear image quality, which can meet the requirements of nano-UAV visual scene dataset construction.
Key words:UAV; Image acquisition system; ZYNQ; Visual scene dataset; FreeRTOS;
收稿日期:2022-07-09
基金项目:国家自然科学基金(62171156)
     下载PDF全文