Abstract: Accurate target detection is the foundation of modern intelligent manufacturing,intelligent transportation and other fields.However,current classic visual foreground extraction methods struggle to meet real-time and accurate requirements in scenes with slow or sudden lighting changes.Therefore,an improved moving object detection method that combined Wronskian function with Visual Background extractor (ViBe) was proposed.The background model was built based on the consistency of temporal and spatial samples,and it was initialized using multiple frames with a specific time interval.Then,the Wronskian function was used to judge the linear correlation between the vector of current pixels and the vector of samples in the background model.The times of linear correlation was used to detect whether the current pixels belonged to moving or background pixels.Finally,the background model was updated using the detection results.Qualitative and quantitative experimental results on real-world scenes showed that the proposed method could effectively overcome the influence of multiple disturbances existing in illumination scenes to extract moving objects.It achieved a higher detection rate and more robust performance when compared with other state-of-the-art methods under slow or abrupt illumination change scenes.

Key words: intelligent manufacturing, intelligent transportation, moving object detection, illumination change

摘要： 精准目标检测是现代智能制造业、智能交通等领域的基础,然而,经典的视觉前景提取方法在光照缓慢或者突变场景中难以满足精确和实时的需求。为此,基于朗斯基函数和视觉背景提取（ViBe）方法提出了一种融合改进的光照变化运动目标检测算法。该方法首先通过时空样本一致性原则构建像素空间背景模型,并利用短时间内采集的多帧图像数据构建初始背景模型;然后,通过朗斯基函数判断当前帧像素构成的向量与背景样本向量的线性相关性,基于当前像素构成的向量与背景模型中样本向量线性相关次数判断当前像素点为前景或者背景;最后,根据判断结果更新背景模型。定性定量实验结果表明,该融合模型能够有效地克服复杂光照变化场景下的多种干扰因素,准确获取光照缓慢或突变场景下的运动目标,且性能优于几种流行的光照变化运动目标检测算法。

关键词: 智能制造, 智能交通, 运动目标检测, 光照变化