用于多目标跟踪的物流货物装载状态监测数据

该数据在道路货物运输领域具有重要的应用价值。能够提供实时货物位移追踪，更精确地检测货物倾斜或滑移状态，帮助物流调度员进行装载安全评估。在干线物流运输中具有广泛的应用场景，特别是公路货运车辆、集装箱装卸作业和物流中转站，能够提高运输过程安全性，降低货物破损率，提供毫米级位移监测精度数据收集：
通过车载多视角摄像头采集货车车厢的1280x720@60fps视频流，同步记录车辆IMU数据。每个样本包含原始图像数据（.raw格式）、人工标注的车厢区域坐标（ROI）、经过直方图均衡化的预处理图像（.npy格式）、货物x方向位移量（人工标注）、货物y方向位移量（人工标注）、货物倾斜角（人工标注）。

数据预处理：
对原始图像进行以下处理：
(1) ROI区域裁剪至512x512像素。
(2) 自适应直方图均衡化增强对比度。
(3) 归一化处理至[-1,1]数值范围。
预处理后生成包含时间戳的numpy矩阵数据，维度为(512,512,3)@float32。

模型构建：
采用改进的DeepSORT算法框架，结合YOLOv8检测器构建货物跟踪系统。关键算法公式包含：

状态向量更新方程：x_k = F * x_{k-1} + B * u_k + w_k。其中x_k和x_{k-1}分别表示k时刻和k-1时刻的车厢ROI坐标(cx,cy)，F为状态转移矩阵，u_k为IMU输入的加速度量测，w_k为过程噪声，B为系数。匹配代价计算：C_{ij} = λ_d(1 - IoU(b_i,b_j)) + λ_a|cosθ_i - cosθ_j| 。式中IoU表示检测框交并比，θ为货物倾斜角（度），λ_d=0.6和λ_a=0.4为加权系数，b_i和b_j表示两次检测结果。算法输出包含每件的货物x方向位移量（模型预测）Δx,货物y方向位移量（模型预测）Δy（单位：cm）和货物倾斜角（模型预测）Δθ（单位：度），通过卡尔曼滤波器与匈牙利算法实现多目标持续跟踪。跟踪准确率采用IDF1指标评估，位移误差通过人工标注数据和模型预测数据计算得到。

This dataset holds significant application value in the road freight transportation domain. It enables real-time cargo displacement tracking, more accurate detection of cargo tilt or slippage conditions, and assists logistics dispatchers in conducting loading safety assessments. It has broad application scenarios in trunk line logistics transportation, especially for road freight vehicles, container loading and unloading operations, and logistics transfer stations, which can improve transportation safety, reduce cargo damage rates, and collect millimeter-level displacement monitoring precision data:

Data collection: Video streams of 1280x720@60fps of truck compartments are collected via on-board multi-view cameras, while vehicle IMU data is recorded synchronously. Each sample includes: original image data (in .raw format), manually annotated region of interest (ROI) coordinates of the compartment, preprocessed images via histogram equalization (in .npy format), manually annotated cargo displacement in the x-direction, manually annotated cargo displacement in the y-direction, and manually annotated cargo tilt angle.

Data Preprocessing:
The original images are processed as follows:
1. Crop the ROI region to 512x512 pixels.
2. Apply adaptive histogram equalization to enhance contrast.
3. Normalize the pixel values to the range [-1, 1].
After preprocessing, numpy matrix data with timestamps is generated, with dimensions of (512, 512, 3)@float32.

Model Construction:
An improved DeepSORT algorithm framework combined with the YOLOv8 detector is adopted to build a cargo tracking system. Key algorithmic formulas include:
State vector update equation: $x_k = F cdot x_{k-1} + B cdot u_k + w_k$. Here, $x_k$ and $x_{k-1}$ respectively represent the ROI coordinates $(c_x, c_y)$ of the compartment at time $k$ and $k-1$, $F$ is the state transition matrix, $u_k$ is the acceleration measurement input from the IMU, $w_k$ is the process noise, and $B$ is the coefficient matrix.
Matching cost calculation: $C_{ij} = lambda_d(1 - IoU(b_i, b_j)) + lambda_a|cos heta_i - cos heta_j|$. In this formula, $IoU$ denotes the intersection over union of detection boxes, $ heta$ is the cargo tilt angle (in degrees), $lambda_d=0.6$ and $lambda_a=0.4$ are the weighting coefficients, and $b_i$ and $b_j$ represent two sets of detection results.
The algorithm outputs the model-predicted cargo displacement in the x-direction $Delta x$, model-predicted cargo displacement in the y-direction $Delta y$ (unit: cm), and model-predicted cargo tilt angle $Delta heta$ (unit: degree) for each cargo item. Multi-objective continuous tracking is implemented via the Kalman Filter and the Hungarian Algorithm. The tracking accuracy is evaluated using the IDF1 metric, and the displacement error is calculated by comparing the manually annotated data and the model-predicted data.

搜集汇总

背景与挑战

背景概述

该数据集由温岭市天航物流有限公司提供，包含5874条CSV格式的企业数据，主要用于多目标跟踪的物流货物装载状态监测。数据通过车载多视角摄像头和IMU设备采集，包含货物位移和倾斜角的人工标注与模型预测结果，应用于道路货物运输领域，可提高运输安全性和监测精度。