视频编码参数对目标识别性能影响的研究

Video Coding Parameters Effect on Object Recognition

国内外研究人员对图像目标分类识别和视频编码传输问题都分别进行了大量研究,但是对于视频编码参数对目标识别性能影响的定量关系,还没有公开的文献报导。针对这一问题,该文选择典型的目标识别算法可变部件模型(DPM)和最常用的视频编码方法 H.264/AVC作用测试对象,通过设计的编码和检测实验,研究了码率和分辨率参数对视频目标识别性能的影响,并拟合了识别性能随码率和分辨率变化的函数关系。通过选取编码器合适的码率和分辨率工作参数,可以获得信道带宽与视频目标识别性能的折中,为设计不同视频应用的编码优化目标函数提供了依据。

Researchers have done a great number of studies on the object recognition and the video coding transmission respectively. However, there are still no public reports about the influence on the object recognition raised by the video encoding parameters. For this issue, the Deformable Part Model （DPM）, a typical object recognition algorithm and the most commonly-used video encoding methods-H.264/AVC are chosen as the test objects. In order to study how the code rates and the resolution affect the performance of video object recognition, the coding and detection experiments are designed and the function of recognition performance changes caused by the code rates and the resolution is fitted. The result shows that the compromise can be achieved between the channel bandwidth and the video object recognition performance through selecting the appropriate the code rates and the resolution parameters for the encoder which provides basis for encoding optimization object function of different video applications.