车辆行驶产生的颠簸、晃动, 会严重影响红外激光车载云台采集视频的质量, 不利于信息的观察和判读。为了改善视频质量, 提出一种实时有效的车载电子稳像方法。首先, 用非线性扩散滤波建立尺度空间, 使视频帧在无损精度的前提下突出边缘信息; 然后, 采取图像亮度信息与梯度信息相结合的特征提取策略,利用图像亮度信息快速搜索潜在特征点并记录其位置, 并通过分析其梯度信息与预判条件的对比结果来选取优质特征点, 在特征描绘阶段, 通过增强二值特征描绘器之间的显著性和差异性, 提升特征量的辨别力; 最后, 通过位置验证辨别相似特征, 提高全局运动矢量估计精度。时间比对实验表明提出的算法能够满足实时处理的要求,在分辨率为720 P时也能达到每秒处理帧频大于60帧; 有效能力对比实验中, 新方法的重复度均高于65%, 明显提高了特征探测能力, 并且在加入去抖动设计后6组实验的帧间转换精度分别提升了46.8%、30.8%、28.44%、28.1%、33.9%和53.4%, 表明该方法极大地改善了抖动视频的稳定性和视频信息提取的准确度。

The quality of vehicle records was seriously influenced along with the bumping and shaking driving, and the jittering videos would impact observer reading and interpreting the information. In order to improve the quality of the records, a real-time image stabilization method for vehicle cloud platforms had been proposed. Firstly, the nonlinear filter was adopted to build scale space for the purpose of highlighting edge information; Secondly, a feature detection combined fast image brightness test and gradient calculation was put forward for more quality features, furthermore, increasing the self-salience and differences between descriptors brought discriminative power; Finally, similar features was availably distinguished by location verification for accurate estimation of global motion vector. The time experimental results shown that the proposed approach fulfile the task of real-time, and the average frame rate is over 60 even when the resolution is 720 P. In effective capability experiments, the repeatability of new method is more than 65%, which indicates the enhancement of detection power. Furthermore, the ITF of 6 group tested increase 46.8%, 30.8%, 28.44%, 28.1%, 33.9% and 53.4% after dejitter design, which illustrates that the method significantly improves the effectiveness and precision of the vehicle image stabilization algorithm.