introduction
H.264 is a new generation of video coding standards developed by ISO (International Standardization Organization) / IEC (International Electrotechnical Association) and ITU (International Telecommunication Standardization Department), which is currently in the application development phase. H.264 has higher compression efficiency and better network adaptability, under the same image quality, H.264 is about 36%, H.263 5L%, MPEG-4 61% U-31. With the improvement of efficiency, the complexity of the algorithm has also increased much, so it is necessary to study the compression algorithm of H.264 and its embedded application.
1 Video encoder hardware design
1.1 PXA255 processor features
Intel Xscale PXA255 Spike is an Intel PCA processor based on Intel Xscale microformors. The internal use of xscale core architecture, with a frequency of up to 400MHz, and kernel expansion, many DSP instructions greatly improve multimedia processing power, and provides an LCD (LCD monitor) controller. , Wireless local area network (802.1ib), Bluetooth, high speed infrared, USB and other functional modules. The PXA255 processor also uses BranchTargetBuffer technology to improve the efficiency of the instruction execution (pipeline). As the preferred embedded processor of the portable mobile device, the PXA255 core has Intel's most advanced dynamic power management technology, which can dynamically adjust the operating frequency and a variety of power save modes according to the task of the processor, suitable for embedded video acquisition systems. application.
1.2 Hardware structure of the system
The hardware structure of the video encoder is shown in Figure 1. The original video stream output from the camera enters the PXA255 chip through the USB interface for H.264 compression. The compressed code stream is transmitted by the network port to the host. The embedded Linux system and application are burn it in the flash. SDRAM Some temporary data used to store the encoding process.
Be
Be
2 realization of video acquisition
Using the VIDE041Inux kernel module in the Linux operating system for video operations, it provides a range of API functions for the video device. This article is to utilize the API function provided by Videalinux to implement video acquisition. The video capture process is shown in Figure 2.
Be
Be
The implementation of the specific video acquisition is as follows:
First invoke a function fd = Open ("DCV / Video", O to open the camera RDWR) device, and then return to the file descriptor FD of the camera. Then by calling the VI.
DCO_GET CAPABILITY0 and VIDEO_GET PICTUREO two functions to get basic information of the camera and the image. Set the brightness, contrast, chrominance and other information of the image using the ICTOL (FD, VidiocgPict, & grab Pic.Ture). Then call the function ICTOL (FD, V1DIOCMCAPTRUE, & GRAB to start capturing a frame image ,.bur) Recall the function ICTOL (FD, VIDIOSYNC, & GRAB BUF) to determine whether this frame image capture ends. This continuously caught multi-frame diagram. The collection of raw video information is completed. Figure 3 is a frame of raw image.
Be
Be
3 H.264 encoding algorithm
H.264 Coding criteria are currently proposed by the latest video compression standards. Compared to previous compression criteria, H.264 further increasing coding efficiency and image quality by modifying an algorithm such as intra prediction, frame prediction, transform coding, and entropy coding.
These improvements include:
(1) When the motion estimation, the size of the macroblock (MB) can be flexibly selected. On the division of macroblocks, H, 264 use 4 modes of 16x6, 16x8, 8x16, 8x8; when divided into 8x8 mode, the 3 seed macroblock division mode of 8x4, 4x8, 4x4 (Figure 4) can be further employed. As shown, further division, doing so that the division of the moving object can be more accurate, reduce the connection error of the edge of the moving object, but also reduce the amount of calculation during the conversion process.
Be
Be
(2) 1/4 pixel precision of motion valuation. The predicted value of the L / 2 pixel position is obtained through the 6th-order FIR filter in H.264. When the L / 2 pixel value is obtained, the value of the L / 4 pixel position is obtained by taking a numeral pixel position and an L / 2 pixel position pixel value mean, which is divided into the inter prediction error, and the transform and quantization are reduced. The number of non-zero bits have improved coding efficiency.
(3) Multi-reference frame motion valuation. Conventional encoding techniques were performed on the P frame (field) image, only previous I frame (field) image or P frame (field) image is allowed to be a reference frame. The previous two I frames (field) images or P frames (field) images are allowed to be referenced images when predicting the B image. H.264 allows one (P prediction mode) or two (B prediction modes, images as a reference image) in multiple images in ReferenceBuffer. The reference image can even be an image using a two-way predicted encoding method.
(4) Selection of reference images is independent of its encoding mode. Allows images with the current image more match to predict the reference image, reduce the prediction error, and improve the coding efficiency.
(5) More accurate intra predictions. In H.264, each pixel in each 4 * 4 can perform intra prediction with different weighting and to which the previously encoded pixels are used.
(6) loop go to block filter. H.264 / AVC Predicting the Demolve Filter Introduction Motion Estimation Prediction Loop, which can remove both square effects, and protect the image detail edge, and also improve the main, objective assessment of the image. Moreover, the filtered image is used in the cache in the cache as needed to further improve the prediction accuracy.
(7) H.264 uses a unified cross-length coding country VIZ) code table. The previous standard entropy coding usually uses the length of Hafman coding, and its code table is not uniform, and it cannot adapt to the change of the multi-end video content, which affects the increase in coding efficiency. Here, that is, the VLC of the H.263 different coefficients is improved, and IRVICs with a unified code table are used, and the VCL method in H.26L is improved, and the quantified DCT transformation is made. The coefficient uses content-based adaptive variable length encoding (CAV), and therefore also defines adaptive binary arithmetic coding (CABAC) based on context content, which is better than CAVLC. Thus, compression coding efficiency is improved by means of UVLC, CAVLC and CABAC.
Implementation of H.264 encoding algorithm on PXA255
ITU-T (International Telecommunication Standardization Department) provides a core algorithm of H.264, which is used herein, which is the open source encoding software X264. Transplant it to the PXA255 embedded development platform after cross-compiling. Since the H.264 encoding algorithm has high complexity, the amount of operation is high, the H.264 coding efficiency is not high, and the preliminary test coding rate is 10 frames per second of QCIF format, and the real-time video coding is not reached.
Therefore, it is necessary to optimize the encoding program. Technology area
Alexa voice service software extension STM32Cube
Combined with DNN API drivers to drive future neural network applications
Four integration design for high-end mobile communications and automotive markets
A frame capture open source graphic debugger application design
The latest security technology that is about to be standard configuration: Rear view camera system
Our other product:
