STCrowd

我们介绍了一个大规模的多模式数据集STCrowd。具体地，在STCrowd中，平均每帧总共有219 K个行人实例和20个人，具有不同的遮挡水平。我们提供同步的LiDAR点云和相机图像以及它们相应的3D标签和关节id。STCrowd可用于各种任务，包括仅激光雷达，仅图像和基于传感器融合的行人检测和跟踪。我们为大多数任务提供基线。此外，考虑到行人的稀疏全局分布和密度变化的局部分布的特性，我们进一步提出了一种新颖的方法，即密度感知分层热图聚合 (DHA)，以增强拥挤场景中行人的感知能力。广泛的实验表明，我们的新方法在各种数据集上达到了行人检测的最新性能。

We introduce a large-scale multimodal dataset, STCrowd. Specifically, STCrowd contains an average of 219 thousand pedestrian instances and 20 individuals per frame with varying levels of occlusion. We provide synchronized LiDAR point clouds, camera images, along with their corresponding 3D labels and joint IDs. STCrowd supports a variety of tasks, including LiDAR-only, image-only, and sensor fusion-based pedestrian detection and tracking. We provide baselines for most of these tasks. Furthermore, considering the characteristics of sparse global distribution and locally varying density distribution of pedestrians, we further propose a novel method named Density-Aware Hierarchical Heatmap Aggregation (DHA) to enhance pedestrian perception in crowded scenes. Extensive experiments demonstrate that our proposed method achieves state-of-the-art performance for pedestrian detection across various datasets.

搜集汇总

背景与挑战

背景概述

STCrowd是一个大规模多模态行人感知数据集，包含同步的LiDAR点云和相机图像，适用于多种行人检测和跟踪任务，并提供了基线方法和新的密度感知分层热图聚合方法(DHA)。