Video-Based Fall Detection and Medical Image Denoising Methods

医疗技术可以帮助管理不断增长的老年人口给医疗体系带来的越来越大的压力，老年人口容易摔倒。图像处理算法构成了跌倒检测系统的基础，可以提高医学图像的质量，从而正确评估伤害的影响。研究人员最近将重点放在了基于视频的跌倒检测解决方案上。文献中提到的方法受到速度和精度的权衡，因为精度的任何改进都会导致更长的执行时间，并且也容易出现误报。该项目寻求调查是否以及如何开发出一种更准确、不牺牲速度的方法。然后，必须在类似真实世界条件的环境中对该方法进行验证。坠落发生后，医生必须对受影响的区域进行扫描，以评估受伤的严重程度。由于硬件引入的噪声和各种伪影，使得医学图像质量下降。对医学图像中的噪声抑制方法进行了广泛的研究。然而，通常只检查一种类型的人工产物。因此，该项目将研究是否以及如何制定一种适应性方法，以确定并有效地对抗多种影响。然后，明智的做法是调查该方法在多大程度上适用于所有类型的医学图像。对于这两个研究部分，在完成文献审查后，将制定算法的初始版本，并收集显示不同程度骨折的坠落场景和医学扫描的数据集。然后将编写机器学习代码来实现和测试算法，以便进行进一步的优化。考虑到数学在信号处理领域的基础作用，建议的研究项目主要可以归类为与数学科学有一些交叉的工程领域。博士项目的价值是多方面的。首先，由于代码将免费提供，因此可以将拟议的跌倒检测算法集成到商业监测解决方案中。因此，一旦发生跌倒事件，护理人员和紧急服务人员将得到迅速通知，以便患者得到及时的医疗救治。此外，医学图像增强方面的工作将通过简化诊断损伤严重程度的过程来帮助医生进行工作。第二个研究部分的研究结果也将对更广泛的研究界有所帮助。了解所提出的方法在多大程度上可以应用于不同类型的医学图像，将使研究人员能够提取与他们自己的工作相关的结论。总而言之，拟议的博士项目将侧重于开发一种健壮的跌倒检测算法和一种自适应医学图像增强技术，目的是改善已经开展的工作和为患者提供的结果。

Medical technology can help manage the growing strain imposed on healthcare systems by the rising elderly population, who is prone to falls. Image processing algorithms form the basis of systems built for fall detection and can enhance medical images' quality, so that the impact of injuries is correctly assessed. Researchers have recently focused on video-based solutions for fall detection. The approaches mentioned in the literature suffer from the speed-accuracy trade-off, as any improvements in accuracy lead to longer execution times, and are also prone to false positives. The project seeks to investigate if and how a more accurate method, which does not sacrifice speed, can be developed. The approach must then be validated in an environment resembling real-world conditions. After a fall has occurred, physicians must use scans of the affected region to assess the injury's severity. Medical images are degraded by noise and various artifacts introduced by hardware. Extensive work has been undertaken on noise suppression methods in medical images. However, only type of artifact is typically examined. The project will hence investigate if and how an adaptive approach, which can identify and effectively combat multiple effects, can be formulated. It would then be wise to investigate to what extent the method is applicable to all types of medical images. For both research components, after the completion of the literature review, an initial version of the algorithms will be formulated and datasets of fall scenes and medical scans showing varying degrees of bone fracture will be collected. Machine learning code will then be written to implement and test the algorithms, so that further optimisations can be made. The proposed research project can be classified primarily into the area of engineering with some cross-over with mathematical sciences, considering mathematics' fundamental role in the signal processing field. The PhD project's value is multifold. Firstly, it will be possible to integrate the proposed fall detection algorithm into commercial monitoring solutions, as the code will be freely available. Carers and emergency services will thus be quickly notified about a fall event, so that the patient receives medical attention in a timely manner. In addition, the work on medical image enhancement will assist physicians with their work by simplifying the process of diagnosing an injury's severity. The findings of the second research component will also be of use to the wider research community. Understanding to what extent the proposed approach can be applied to different types of medical images will allow researchers to extract conclusions relevant to their own work. To summarise, the proposed PhD project will focus on the development of a robust fall-detection algorithm and of an adaptive medical image enhancement technique with the aim of improving the work already undertaken and outcomes for patients.