针对固定监控场景提出了一种运动目标检测与跟踪方案。在运动目标检测中,利用像素梯度及色度均值、方差分布建立并实时更新背景模型。在目标跟踪模块,引入卡尔曼滤波器预测目标参数,合并目标碎片,建立帧间目标匹配矩阵完成目标匹配。通过实际图像序列测试,算法能较好地实现运动目标跟踪,获得运动目标的轨迹,具有良好的实时性和适应环境变化的能力。
A method of moving object detection and tracking in stationary scene is presented. The background model based on the mean and variance of gradient and chromaticity is using for detecting objects. In tracking module, the matrix using for recognition is built between two frames. Kalman filter is used for predicting objects' parameters, merging objects' fragments, and its predicting parameters is provided for matching objects. The method has been tested on image sequence to show the validity of the approach.