Abstract: If using the traditional algorithms of background extraction and update on the fixed-point DSP, which is memory-limited and weak at floating-point operation, the detection speed is slow and the real-time ability is poor. The algorithm initializes the background reference frame according to the hypothesis that the background pixel intensity appears in video sequence with maximum probability and the temporal relativity between two neighboring frames. In the session of background update, the algorithm updates the background reference frame by assigning different weights to the original background and the current frame and calculating the result based on shifting operation. It is successfully implemented on TMS320DM642 and results show that the algorithm can meet the real-time constraint in 25 FPS.

Key words: background extraction and update, fixed-point DSP, real-time ability

摘要： 定点DSP浮点运算效率低、存储空间有限，应用传统的背景提取与更新算法时，检测速度慢、实时性差。针对该问题，利用相邻视频帧的时间相关性，根据背景像素灰度以最大概率出现在视频序列中的假设，提取初始背景参考帧。在背景更新环节，对原始背景与当前图像像素赋予不同的权值，并且基于移位运算得到新的背景。在TMS320DM642定点DSP上所做的实验表明，该算法可以满足帧率为25 FPS下的实时应用。

关键词: 背景提取及更新, 定点DSP, 实时性

CLC Number: 

• TP391.41