近年来，目标检测技术因其能够对视频和图像中的目标进行定位和分类而受到了广泛的关注。传统的目标检测算法需要大量的计算资源来建立模型。人们越来越需要一种实用的方法来构建适合本地上下文、有限计算资源和应用逻辑的目标检测模型，同时支持实时推断而不降低准确性和性能。

在本论文中，我们提出了一个分布式的合作架构来建立实际的目标检测模型。该框架基于一种新的协作群体推理方法，该方法采用单类单模型机制对多个对象进行分布式推理。为了进行有效的分组，我们在推断过程中利用了现有模型中的组内相关性。从Pascal VOC 2007的案例研究和我们通过Google Street View收集数据中得到的结果表明，所提出的模型显著地提高了性能，同时使其有可能适用于计算资源有限的定制应用程序构建。建立了该模型的原型系统，并通过一个邻域应用实例验证了该模型的有效性。

Object detection has gained much attentionin recent years because of its ability to localize and classify the objects invideos and images that can be incorporated into many applications. Traditionalobject detection algorithms need substantial computational resources to build amodel. There is an increasing demand for a practical approach to constructingobject detection models adapted to the local context, limited computingresources, and application logics while supporting real-time inferencingwithout degrading accuracy and performance. In this thesis, we proposed adistributed-collaborative framework to build practical object detection models.The framework is based on a novel approach for collaborative group inferencingthat is designed with a single-class-single-model mechanism for multipleobjects in a distributed manner. For useful grouping, we made use of theintraclass correlation from existing models during inferencing. Results fromthe case studies with Pascal VOC 2007 and our data collected through GoogleStreet View showed that the proposed model significantly improved performancewhile making it potentially suitable for customized application building withlimited computing resources. A prototype for the proposed model has been built,and a neighborhood application has been demonstrated to validate the proposedwork.

1.  引言

2. 项目背景与相关工作

3. 研究方法

4. 实验与结果

5. 结论与展望

更多精彩文章请关注公众号：