Dataset - CQU_TMR

Road safety systems are essential for planning, managing, and improving road infrastructure and decreasing road accidents. Manual systems used for road safety assessments are inefficient, time consuming, and prone to error. Some automated systems using sensors, cameras, lidar, and radar to detect nearby obstacles such as vehicles, pedestrians, lane lines, some traffic signs and parking slots have been introduced to reduce road fatalities by minimizing human error. However, the existing road safety systems available in industry are unable to accurately detect all road safety attributes required by the Australian Road Assessment Program (AusRAP), a program launched to establish a safer road system through high-risk roads inspection, developing star ratings and safer roads investment plans to mitigate the possibility of meeting with accidents. Therefore, it is important to explore novel techniques and develop better automated systems which can accurately detect and classify all road safety attributes.

This research focuses on the development of a novel deep learning technique for the analysis of road safety attributes. Various architectures, learning and optimisation techniques have been investigated to develop an appropriate deep learning-based technique that can detect road safety attributes with high accuracy. Firstly, a single-stage segmentation and classification technique to automatically identify AusRAP attributes has been investigated. Secondly, multi-stage segmentation and classification techniques using various classifiers have been investigated. Finally, Genetic Algorithm (GA) and Particle Swarm Optimisation (PSO)-based techniques have been investigated to optimise the proposed deep learning techniques.

The proposed techniques were evaluated on a real-world dataset using roadside videos provided by the Department of Transport and Main Roads (DTMR), Queensland, Australia, and Australian Road Research Board (ARRB). The classification accuracy was used as a metric to measure the performance, and to further validate the efficacy, different diversity measures such as specificity, sensitivity, and f1-score were used. An appropriate analysis and a comparison with existing techniques were conducted and presented. The results and analysis show that the proposed single-stage and multi-stage deep learning-based techniques achieve classification accuracy and misclassifications better than the existing state-of-the-art segmentation and classification techniques. It was found through experimentation that proposed single stage technique can avoid re-training the whole model using all training samples which requires a lot of time when a new attribute is introduced. Moreover, through extensive experimentation, it was found that it is not always necessarily required to have a large dataset for training. Effective solutions were found to eliminate the requirement to annotate large number of samples for each attribute to produce acceptable accuracy for industry. Both single-stage and multi-stage deep learning-based techniques were also validated using real world test data without cropping and pixel wise prediction was obtained for each object. The accurate location of the predicted object was known in predictions and hence, bounding box problem was avoided. Through the incorporation of optimisation techniques, optimum parameters suitable for road safety attributes were determined. The optimum parameters proved to be effective in terms of classification accuracy and time to achieve minimum error.

道路安全系统对于道路基础设施的规划、管理与优化，以及降低道路交通事故发生率至关重要。传统的人工道路安全评估系统效率低下、耗时耗力且易出现误差。当前已有部分自动化系统借助传感器、相机、激光雷达（lidar）与雷达等设备，对车辆、行人、车道线、部分交通标志及停车位等周边障碍物进行检测，旨在通过减少人为失误降低道路死亡事故的发生。然而，当前工业界现有的道路安全系统，无法精准检测澳大利亚道路评估计划（AusRAP）所要求的全部道路安全属性。该计划旨在通过高风险道路巡检、制定星级评级标准及安全道路投资方案，以降低交通事故发生概率。因此，探索新型技术、研发可精准检测与分类全部道路安全属性的自动化系统具有重要意义。

本研究聚焦于研发用于道路安全属性分析的新型深度学习技术。为开发出可高精度检测道路安全属性的适配性深度学习方案，团队对多种网络架构、学习方法及优化技术展开了研究。首先，研究了用于自动识别AusRAP属性的单阶段分割与分类技术；其次，探究了基于多种分类器的多阶段分割与分类技术；最后，研究了基于遗传算法（GA）与粒子群优化（PSO）的技术，以对所提出的深度学习方案进行优化。

本研究采用澳大利亚昆士兰州交通与主干道部（DTMR）及澳大利亚道路研究委员会（ARRB）提供的道路监控视频，在真实世界数据集上对所提出的技术进行了评估。以分类准确率作为性能衡量指标，并借助特异度、灵敏度、F1分数（f1-score）等多种多样性评估指标进一步验证方案的有效性。研究开展了针对性分析，并与现有技术进行了对比。结果与分析表明，所提出的单阶段与多阶段深度学习技术在分类准确率与误分类率方面均优于当前主流的分割与分类技术。实验发现，所提出的单阶段技术可避免在新增属性时，使用全部训练样本对整个模型进行重新训练（该过程往往耗时极长）。此外，通过大量实验可知，模型训练并非始终需要大规模数据集。研究找到了有效的解决方案，可免除为每个属性标注大量样本以达到工业界可接受的准确率的需求。单阶段与多阶段深度学习技术均通过未经过裁剪的真实测试数据完成了验证，可对每个目标进行逐像素预测，且在预测过程中可获取目标的精确位置，因此规避了边界框问题。通过引入优化技术，确定了适配道路安全属性的最优参数，该最优参数在提升分类准确率与降低误差耗时方面均表现出色。