【论文详解】In Defense of Classical Image Processing: Fast Depth Completion on the CPU

论文《In Defense of Classical Image Processing: Fast Depth Completion on the CPU》提出了一种基于经典图像处理的快速深度补全算法,无需训练数据和深度学习模型。该算法在CPU上实现90Hz频率,效果在KITTI benchmark中领先。包括深度反转、自定义核膨胀、小孔闭运算等8个步骤,以填充稀疏深度图。然而,对于不同稀疏分布的数据,算法效果可能受限。

摘要生成于 C知道 ,由 DeepSeek-R1 满血版支持, 前往体验 >

        论文《In Defense of Classical Image Processing: Fast Depth Completion on the CPU》提出了一种用经典的图像处理算法进行深度补全的算法,不依赖于任何训练数据,没有使用深度学习模型。

主要贡献:

1.提出了一种快速的深度补全的算法,能从稀疏的深度图中恢复完整的深度图,并且在CPU上可达到90Hz的频率。该算法的效果在KITTI depth completion benchmark中排名第一(公布时);

2.该算法的表现超越了其他现有的CNN算法。

本文提出的算法一种包含8个步骤:

(1)Depth Inversion:对于稀疏图像的处理机制是应用OpenCV的形态学操作,用较大的像素值覆盖较小的像素值。数据集KITTI raw depth map,深度值范围在0~80米之间,没有有效深度值的位置用0填充。在这种情况下,如果直接应用膨胀操作会使大距离覆盖掉小距离,从而导致较近对象的边缘信息丢失。为了解决这个问题,将有效深度做一个invert,即D(inverted) = 100.0-D(input),这样在有效深度和空值之间有一个20m的buffer。

(2)Custom Kernel Dilation(用自定义的核进行膨胀操作):先填充最接近有效像素值的空值,关键在于kernel形状的设计,要使得最可能具有相同深度值的像素被填补。因此在这一步采用的是5*5的菱形kernel,形状如下图:

(3)Small Hole Closure(对小的空洞进行闭运算):经过第一步的膨胀操作后,仍任有一些小孔,由于这些区域没有有效深度值,考虑到物体的结构,发现膨胀后相邻的深度可以连接起来形成物体的边缘。因此在这一步用

The classical pipeline of data processing typically consists of several stages or phases. These phases are: 1. Data Collection: This is the first stage of the pipeline where data is collected from various sources such as databases, web pages, APIs, and other sources. The intention of this phase is to gather data that is relevant to the problem being solved. 2. Data Pre-processing: In this stage, the collected data is cleaned, transformed, and organized in a structured format. The intention of this phase is to ensure that the data is consistent and ready for analysis. 3. Data Analysis: In this stage, the pre-processed data is analyzed using various statistical and machine learning techniques to extract valuable insights. The intention of this phase is to identify patterns, trends, and anomalies in the data. 4. Data Visualization: In this stage, the analyzed data is visualized using graphs, charts, and other visual aids. The intention of this phase is to communicate the insights and findings to stakeholders in an easy-to-understand format. 5. Decision Making: In this stage, the insights and findings obtained from the previous stages are used to make informed decisions. The intention of this phase is to take actions based on the insights and findings to improve business processes or solve the problem at hand. Overall, the classical pipeline of data processing is intended to turn raw data into actionable insights that can be used to improve business processes, solve problems, and drive innovation.
评论 4
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值