目标计数作为一项基础的技术,在许多领域都有广泛的应用,如人群计数、细胞计数、车辆计数等。随着互联网时代的信息爆炸,视频数据呈指数级增长,如何快速、准确地获得目标的数量是用户普遍关心的主要问题之一。得益于计算机视觉技术的快速发展,基于传统机器学习的计数方法正逐步向基于深度学习的方法转变,并在计数的准确性上取得了实质性的进展。介绍了目标计数的研究背景和应用领域,根据模型任务分类,归纳了三类常用的计数模型框架,并从不同的角度分别介绍了近10年来基于计算机视觉技术的模型方法。然后介绍了在人群计数、细胞计数和车辆计数领域中常用的几种公开数据集,并横向比较了各个模型之间的性能。最后总结了现阶段的目标计数模型还存在的不足,并对未来的研究方向进行了展望。

Total green leaf area (GLA) is an important trait for agronomic studies. However, existing methods for estimating the GLA of individual rice plants are destructive and labor-intensive. A nondestructive method for estimating the total GLA of individual rice plants based on multiangle color images is presented. Using projected areas of the plant in images, linear, quadratic, exponential and power regression models for estimating total GLA were evaluated. Tests demonstrated that the side-view projected area had a stronger relationship with the actual total leaf area than the top-projected area. And power models fit better than other models. In addition, the use of multiple side-view images was an efficient method for reducing the estimation error. The inclusion of the top-view projected area as a second predictor provided only a slight improvement of the total leaf area estimation. When the projected areas from multi-angle images were used, the estimated leaf area (ELA) using the power model and the actual leaf area had a high correlation coefficient (R2 > 0:98), and the mean absolute percentage error (MAPE) was about 6%. The method was capable of estimating the total leaf area in a nondestructive, accurate and efficient manner, and it may be used for monitoring rice plant growth.

谷物粒型是决定谷粒品质和产量的重要参数之一。传统人工测量粒型的方法耗时、工作量大、主观性强。本文首先介绍一种基于线阵列采集技术和工业输送技术的谷物粒型自动测量系统。为提高系统测量效率，文章中应用了图形处理器(GPU)并行处理技术，在统一计算设备架构(CUDA)下对测量算法进行优化。实验结果表明，基于GPU 的并行加速算法，能有效提高测量效率，当图像中谷粒数近2 000 颗时，优化后的算法速度为中央处理器(CPU)下算法运行速度的400 多倍，且随着采集图像中谷粒数的增多，优化测量算法的加速效果更显著。