The visual system uses two complimentary strategies to process multiple objects simultaneously within a scene and update their spatial positions in real time. It either uses selective attention to individuate a complex, dynamic scene into a few focal objects (i.e., object individuation), or it represents multiple objects as an ensemble by distributing attention more globally across the scene (i.e., ensemble grouping). Neural oscillations may be a key signature for focal object individuation versus distributed ensemble grouping, because they are thought to regulate neural excitability over visual areas through inhibitory control mechanisms. We recorded whole-head MEG data during a multiple-object tracking paradigm, in which human participants (13 female, 11 male) switched between different instructions for object individuation and ensemble grouping on different trials. The stimuli, responses, and the demand to keep track of multiple spatial locations over time were held constant between the two conditions. We observed increased α-band power (9-13 Hz) packed into oscillatory bursts in bilateral inferior parietal cortex during multiple-object processing. Single-trial analysis revealed greater burst occurrences on object individuation versus ensemble grouping trials. By contrast, we found no differences using standard analyses on across-trials averaged α-band power. Moreover, the bursting effects occurred only below/at, but not above, the typical capacity limits for multiple-object processing (at ~4 objects). Our findings reveal the real-time neural correlates underlying the dynamic processing of multiple-object scenarios, which are modulated by grouping strategies and capacity. They support a rhythmic, α-pulsed organization of dynamic attention to multiple objects and ensembles.

SIGNIFICANCE STATEMENT Dynamic multiple-object scenarios are an important problem in real-world and computer vision. They require keeping track of multiple objects as they move through space and time. Such problems can be solved in two ways: One can individuate a scene object by object, or alternatively group objects into ensembles. We observed greater occurrences of α-oscillatory burst events in parietal cortex for processing objects versus ensembles and below/at versus above processing capacity. These results demonstrate a unique top-down mechanism by which the brain dynamically adjusts its computational level between objects and ensembles. They help to explain how the brain copes with its capacity limitations in real-time environments and may lead the way to technological innovations for time-critical video analysis in computer vision.

视觉系统使用两种互补策略来同时处理场景中的多个对象并实时更新其空间位置。它要么使用选择性注意力将一个复杂的动态场景划分为几个焦点对象（即，对象个体化），要么通过在整个场景中更广泛地分配注意力来将多个对象表示为一个整体（即，整体分组）。神经振荡可能是焦点对象个体化与分布式整体分组的关键特征，因为人们认为神经振荡通过抑制控制机制调节视觉区域的神经兴奋性。我们在多对象跟踪范例中记录了全头MEG数据，其中人类参与者（13位女性，11位男性）在不同的指令之间切换对象个体化和整体分组的不同指令。在两个条件之间，刺激，响应和随时间推移跟踪多个空间位置的需求保持不变。我们观察到在多对象处理过程中，双侧下顶叶皮层的振荡爆发中增加了α波段功率（9-13 Hz）。单项试验分析显示，与组合分组试验相比，对象个体化的爆炸发生率更高。相比之下，我们使用跨试验平均α频段功率的标准分析发现没有差异。此外，爆发效应仅发生在多对象处理的典型容量限制以下/处，而不是以上（〜4个对象）。我们的发现揭示了多对象场景动态处理基础上的实时神经相关性，这些相关性由分组策略和能力进行调节。它们支持对多个物体和合奏进行动态注意的有节奏的，α脉冲组织。

重要性声明动态多对象方案是现实世界和计算机视觉中的重要问题。它们需要跟踪在空间和时间中移动的多个对象。这些问题可以通过两种方式解决：一种可以逐个对象地将场景对象个体化，也可以将对象分组为整体。我们观察到顶叶皮层中的α振荡震荡事件的发生率相对于整体而言更大，而低于/高于与高于加工能力则更高。这些结果证明了一种独特的自上而下的机制，大脑可以通过该机制动态地调整其在对象和集合体之间的计算水平。它们有助于解释大脑如何在实时环境中应对其能力限制，并可能引领计算机视觉中对时间要求严格的视频分析的技术创新。