Research On Intelligent Video Coding Method Based On Target Detection

With the rapid development of image acquisition and display technology,high-resolution video such as 4K/8K is gradually entering our work and life.Relative to the H.264 standard encoding method,the new generation video compression standard HEVC saves about 50% of the code stream on the high-resolution video coding,but the compressed video data is still large due to the multiplication of the resolution of HD video images,and the current network bandwidth resources are still relatively limited,which brings great challenges to the transmission and storage of video data.The existing video coding standard can effectively remove the spatial redundant information of the video when encoding the video,but does not take into account the visual characteristics of the human eye,for those areas that do not conform to human visual characteristics,it often consumes a lot of bit rate resources,which affects the image quality of those areas of interest.Therefore,how to properly allocate the code rate resource of video coding under limited network bandwidth and storage space is very important to make the compressed video image more in line with human visual characteristics.Therefore,how to reasonably allocate the bitrate resources of video coding under the limited network bandwidth and storage space,so as to make the compressed video image more consistent with human visual characteristics is of great importance.In order to solve the above problems,this paper proposes an intelligent video coding method based on target detection.The main work and achievements of the thesis are as follows:(1)The X265 coding platform is built,the key technologies of HEVC coding standard are analyzed.According to the human visual characteristics,the overall scheme of intelligent video coding method based on target detection is designed,andthe video quality evaluation method is selected.(2)Aiming at the problem that the traditional region of interest coding methods can not detect the specific target when detecting the target of interest,a method of extracting the region of interest based on convolutional neural network is proposed.Firstly,the targets are determined according to the application scenario,the convolutional neural network model is selected and trained.Then the video image is input into the convolutional neural network to detect the targets object of interest,and extracting the location coordinates,the maximum confidence target object category and the number of target objects.Finally,the extracted position coordinates are post-processed to generate a region of interest.(3)According to the characteristics of human visual system,a method of generating visual perception map is designed for subsequent region of interest coding.Firstly,the edge intensity of each pixel is calculated by a high-pass filter,and the direction attribute of each pixel is determined.Then the texture complexity in the current CU block is calculated,and a CTU-level texture sensing map is generated according to the texture intensity information of each CU block.Finally,flat regions,structured texture regions,and complex texture regions are extracted and texture-aware maps are generated.(4)Aiming at the problem of bit resources allocation in video coding process,a HEVC region of interest coding method based on visual perception is proposed.According to the complexity of the coding block,the quantization parameters of the region of interest are adjusted to different degrees,and the high-frequency coefficients are suppressed in the transform domain by the method of suppressing the frequency coefficient in the non-interest region to realize the rational allocation of bit resources.(5)A smart video encoder based on Jetson TX2 is designed and implemented.Firstly,Jetson TX2 is selected as the hardware development platform to design the intelligent video encoder,which mainly includes video acquisition,region of interest extraction,region of interest coding and code stream transmission module.Finally,through the test analysis,the coding effect of the intelligent video encoder is verified,and the intelligent coding function of the region of interest is realized,which improves the image quality of the region of interest.