利用视频图像检测车位状态的方法

为了实现停车场空车位的快速自动寻找,提出了一种用于视频图像车位检测的车位组合状态网络模型,在最大后验概率意义下,求解最可能的车位状态.对改进的HSI车位图像像素进行贝叶斯分类,利用预先训练好的条件概率计算出相邻车位状态传递概率,通过车位组合状态网络的优化,确定出最佳的车位状态.该方法可以很好地解决车辆遮挡所造成的误判问题,准确地检出车位的占有情况.实验结果表明,该方法具有较高的准确性与鲁棒性,特别是可以较好地消除车辆相互遮挡和环境光线变化对检测结果造成的影响.

Approach for parking spaces detection based on video images

To automatically find the vacant parking spaces quickly,a parking spaces status combinatory network model based on MAP for parking spaces detection is proposed.Firstly,the HSI parking space pixels are classified by Bayesian classifiers.Then the transfer probability of the neighboring parking spaces status is calculated by different categories’ probabilities which are pre-trained.In the end,the most ideal status combinatory is assured through the optimization of the parking spaces status combinatory network by the shortest path algorithm.This approach can not only detect the parking spaces accurately,but also resolve the occlusions caused by the neighboring pairs.Experimental results indicate that the algorithm has high veracity and robustness,especially can eliminate the effects caused by luminance variations and occlusions of neighboring pairs.