智能识别非法占用公交专用道算法模型的图像训练数据

本数据集主要用于提升AI模型对社会车辆非法占用公交专用道行为的识别能力与精确性。通过对该数据集的训练，使AI模型能够实现对社会车辆占用、跨线行驶等情况的识别，并可应用于城市公交优先道执法、早晚高峰专用道监管、公交线路监控等场景。同时，本数据集可为交通管理部门提供智能化执法依据，有效提升公交专用道使用监管效率，保障公共交通优先通行权益。
1.数据采集
通过企业自有摄像设备自行采集道路图像，同步记录图像ID、采集时间、设备型号、地理坐标、光照条件、天气状况、车道标线类型、公交车辆在位状态等数据。
2.数据预处理与标注
通过数据清洗，剔除公交车、应急车辆等合法使用专用道的场景，过滤临时交通管制等特殊时段数据。按6:2:2比例划分训练集/验证集/测试集。设置多级标注体系：
​一级标签：合法使用/非法占用
​二级标签：社会车辆全程占用（连续占用≥50米）/社会车辆跨线行驶（车轮压线超过50%宽度）/社会车辆临时停靠（占用时长10-30秒）/社会车辆长时间停靠（占用时长＞30秒）
​辅助标注：车道标线坐标、车辆与专用道位置关系（完全/部分占用）
3.模型选择与初始化
采用YOLOv8s-UNet++混合架构，初始化参数并优化超参数：学习率0.01-0.001动态调整，批量大小1-32动态调整，锚框参数自动调整适配常见车辆形态）；集成坐标注意力机制（CA模块）提升小目标车辆与车道线的关联检测精度。
4.模型训练
基于PyTorch实施分布式训练，采用混合精度训练（FP16）提升效率。设置训练时长，数据增强模拟复杂场景，添加动态模糊、遮挡物干扰等特效，模拟夜间低光照及雨雾天气条件。设置早停机制（patience=10），梯度裁剪：max_norm=1.5。
5.模型评估
在训练模型的过程中，使用验证集调整超参数，训练完成后在测试集上评估模型表现，评估指标包含：
基础性能：mAP@0.5、误报率
场景鲁棒性测试：雨雾天气检出率

This dataset is primarily developed to improve the recognition capability and accuracy of AI models in identifying illegal occupation of bus-only lanes by social vehicles. Training on this dataset enables AI models to recognize scenarios including social vehicles occupying bus lanes and driving across lane lines, and can be applied to scenarios such as urban bus priority lane law enforcement, peak-hour dedicated lane supervision, and bus route monitoring. Furthermore, this dataset can provide intelligent law enforcement references for traffic management departments, effectively enhancing the supervision efficiency of bus-only lane usage and safeguarding the priority passage rights of public transport.

1. Data Collection
Road images are collected independently using the enterprise's own camera equipment, with synchronized recording of relevant metadata including image ID, collection time, device model, geographic coordinates, lighting conditions, weather conditions, lane marking type, and the in-lane status of buses, etc.

2. Data Preprocessing and Annotation
Data cleaning is first conducted to exclude scenes where legally authorized vehicles such as buses and emergency vehicles use the dedicated lane, and filter out data from special periods like temporary traffic control. The dataset is split into training, validation and test sets at a ratio of 6:2:2. A multi-level annotation system is established as follows:
- Primary label: Legal use / Illegal occupation
- Secondary labels: Full-time occupation by social vehicles (continuous occupation ≥ 50 meters) / Lane crossing by social vehicles (wheels covering more than 50% of the lane width) / Temporary parking by social vehicles (occupation duration 10-30 seconds) / Long-term parking by social vehicles (occupation duration > 30 seconds)
- Auxiliary annotations: Lane marking coordinates, positional relationship between vehicles and dedicated lanes (full / partial occupation)

3. Model Selection and Initialization
A hybrid architecture of YOLOv8s-UNet++ is adopted, with parameter initialization and hyperparameter optimization carried out: dynamically adjusting the learning rate within the range of 0.01 to 0.001, dynamically adjusting the batch size between 1 and 32, and automatically tuning anchor box parameters to adapt to common vehicle shapes. The Coordinate Attention (CA) module is integrated to improve the detection accuracy of small target vehicles and their correlation with lane lines.

4. Model Training
Distributed training is implemented based on PyTorch, with mixed-precision training (FP16) adopted to boost training efficiency. A specified training duration is set, and data augmentation is applied to simulate complex scenarios, including adding effects such as dynamic blur and occlusion interference, and replicating low-light nighttime and rainy/foggy weather conditions. An early stopping mechanism with patience=10 is configured, and gradient clipping with max_norm=1.5 is utilized.

5. Model Evaluation
During the model training phase, the validation set is used to adjust hyperparameters. After the completion of training, model performance is evaluated on the test set. The evaluation metrics include:
- Basic performance indicators: mAP@0.5, false positive rate
- Scenario robustness test: Detection rate in rainy/foggy weather

搜集汇总

背景与挑战

背景概述

该数据集包含587条图像训练数据，用于训练AI模型识别社会车辆非法占用公交专用道的行为，如全程占用和跨线行驶。数据每日更新，采用YOLOv8s-UNet++混合架构进行模型训练，可应用于城市公交执法和高峰监管场景，提升监管效率和公共交通权益保障。