帧差分智能视频监控系统图像亮度的校正

在基于帧差分的智能视频监控系统中，图像亮度改变对目标物体的识别和跟踪可能造成严重影响。通过对比一些常用亮度校正方法，选用比例变换法进行深入研究。为保证校正效果，预先指定一个不受目标物体影响的区域，根据其平均灰度确定变换因子。为满足实时性要求，仅当连续2帧的亮度改变超过某一阈值时才进行校正。在Visual C++ 6.0下编程时又采取了一些措施提高程序运行速度，例如将浮点运算转换为整数运算等。实验表明:亮度校正前无法识别出的目标物体在亮度校正后可识别出来，校正一帧图像耗时约1.2 ms，当阈值取1或2时校正效果与校正速度综合最好。

Image brightness correction for frame-difference-based intelligent video monitoring systems

In frame-difference-based intelligent video monitoring systems, changes of image brightness can seriously influence the results of target object recognizing and tracking. After comparing some common brightness correction methods, the direct proportion transformation method was selected for further study. To meet the effect requirement, the transformation factor was calculated according to the average grayscale of a manually preselected region for avoiding influence of target object. To meet the real time requirement, the correction process was started only when the brightness difference between two continuous frames was greater than a certain threshold. When programming with Visual C++ 6.0, some further measures were adopted to speed up the process, such as transforming floating-point calculation into integer calculation. Experiment results show that, target objects that cannot be recognised before correction can be recognised after correction, the time consumption is about 1.2 ms for correction of one frame, and the comprehensive effect and speed reach the best when the threshold is 1 or 2.