作者：Fan Yang， Heng Fan， Peng Chu， Erik Blasch， Haibin Ling

发布时间：2019

发布期刊：ICCV

**论文全称：**Clustered Object Detection in Aerial Images

论文地址：http://openaccess.thecvf.com/content_ICCV_2019/papers/Yang_Clustered_Object_Detection_in_Aerial_Images_ICCV_2019_paper.pdf

代码：https://github.com/fyangneil/Clustered-Object-Detection-in-Aerial-Image

地位：这篇论文被ICCV2019收录。主要作者是Fan Yang，该作者来自于天普大学，主要的研究方向是航拍图像下的目标检测。美国费城天普大学，美国空军研究实验室，美国石溪大学共同提出ClusDet用于航空图像的目标检测。

一、摘要

Detecting objects in aerial images is challenging for at least two reasons: (1) target objects like pedestrians are very small in pixels, making them hardly distinguished from surrounding background; and (2) targets are in general sparsely and non-uniformly distributed, making the detection very ineffificient. In this paper, we address both issues inspired by observing that these targets are often clustered. In particular, we propose a Clustered Detection (ClusDet) network that unififies object clustering and detection in an end-to-end framework. The key components in ClusDet include a cluster proposal sub-network (CPNet), a scale estimation sub-network (ScaleNet), and a dedicated detection network (DetecNet). Given an input image, CPNet produces object cluster regions and ScaleNet estimates object scales for these regions. Then, each scale-normalized cluster region is fed into DetecNet for object detection. ClusDet has several advantages over previous solutions: (1) it greatly reduces the number of chips for fifinal object detection and hence achieves high running time effificiency, (2) the cluster based scale estimation is more accurate than previously used single-object based ones, hence effectively improves the detection for small objects, and (3) the fifinal DetecNet is dedicated for clustered regions and implicitly models the prior context information so as to boost detection accuracy. The proposed method is tested on three popular aerial image datasets including VisDrone, UAVDT and DOTA. In all experiments, ClusDet achieves promising performance in comparison with state-of-the-art detectors.

• 作者发现在遥感图像上的两个难点：
  1. 行人等目标对象的像素很小，很难与周围的背景区分开来
  2. 目标一般是稀疏分布和不均匀分布的，这使得检测效率非常低
  • 难点一的解决方案就是将原图分块，难点一的解决方案则带来了一些图像块没有对象，会浪费计算资源，因此作者提出了使用基于聚类的分块，其对比可由此处查看
• 作者受到行人等目标对象有聚集的特性，提出了Clustered Detection（ClusDet），是一个统一了对象聚类和检测的端到端框架。主要由三部分组成：
  1. cluster proposal sub-network（CPNet）
  2. scale estimation sub-network（ScaleNet）
  3. dedicated detection network（DetecNet）
  • 总体流程：输入图像 → CPNet产生对象聚集区域 → ScaleNet估计这些区域的对象尺度 → 将每个尺度归一化聚类区域输入到DetecNet进行目标检测
• Clustered Detection（ClusDet）的优点：
  1. 它大大减少了最终目标检测的图像块数量，从而实现了较高的运行时间效率
  2. 基于聚类的尺度估计比以往使用的基于单目标的尺度估计更准确，从而有效地提高了对小目标的检测
  3. 最终的DetecNet专门用于集群区域，并隐式地对上下文信息建模，以提高检测精度
• 在VisDrone，UAVDT和DOTA数据集上进行实验

二、Clustered Detection

2.1 整体网络架构

• 该网络主要由三部分组成：
  1. cluster proposal sub-network（CPNet）：根据对航空图像的初始检测，CPNet生成一组目标簇区域。在获得聚类区域后，将其裁剪出来以进行后续的精细检测，因此，这些区域必须首先调整大小以适应检测器，这可能会导致在集群区域中产生极大或极小的物体，从而降低检测性能（为了解决这个问题，提出了ScaleNet）
  2. scale estimation sub-network（ScaleNet）：估计每个集群区域中的对象的适当规模，然后在将区域喂到检测器之前格局对象的规模重新缩放区域的大小
  3. dedicated detection network（DetecNet）：所有集群区域都被喂到该网络做精细检测。
• 最终的检测结果是通过融合集群区域和全局图像上的检测结果来实现的。