Background Updating Technique in Complex Traffic Scene Based on Sensor Fusion

The challenges of constructing complex traffic background discourage the application of video surveillance approaches. In order to solve the problems, a novel background updating method based on sensors fusion is proposed. First of all, A Gaussian mixture model based on Expectation-Maximization (EM) algorithm is proposed to improve the accuracy of background modeling. Secondly, for slowly moving or stationary region of interest, different sample frequencies are set to increase the reliability of modeling data. Meanwhile, data from loop and virtual loop fusion is performed to enhance the accuracy of vehicle speed detected by virtual loop. A local stationary optimal Kalman smoother is introduced in order to improve the self-tuning ability of the data fusion algorithm. Finally, experimental results prove the validness and robustness of the method.

摘要复杂交通场景由于背景构造困难限制了视频监控技术的应用。为了解决这个问题，本文提出了一种新的基于传感器融合的复杂交通视频背景构造方法。首先通过基于EM算法的高斯混合模型提高背景建模的准确性；其次，对于运动缓慢或者静止的区域，通过设定不同的采样频率来提高背景建模数据的可靠性；对来自线圈和虚拟检测线检测到的数据进行数据融合来提高虚拟线圈检测车速的精度，通过引入局部稳态最优Kalman平滑器提高了数据融合算法的自校正能力。最后，实验结果证明了所提方法的有效性。