基于局部聚合描述符的视点不变视觉位置识别

View-invariant visual position recognition based on vector of locally aggregated descriptors

针对目前视觉位置识别系统无法同时拥有视点不变、条件不变和高效率的性能,提出一种基于视点不变的位置识别系统,采用局部聚合描述符(VLAD)。采用高稳健性的加速稳健特征(SURF)算子进行特征检测,利用梯度直方图描述特征;利用VLAD将每个特征划分到特定的聚类中,使用局部敏感散列(LSH)进行数据降维,将特征随机投影到低维的二进制签名中,计算二进制签名;通过计算二进制图像签名上的汉明距离实现图像比较,识别视觉位置。实验结果表明,所提系统只需要占用很小的内存,就能够稳健适应变化的环境。

In view of the fact that the current visual position recognition system can not simultaneously have the performances of view invariance,condition invariance and high efficiency,a view invariant position recognition system was proposed,in which the vector of locally aggregated descriptor(VLAD)was used.The speeded-up-robust-features(SURF)operator was used to detect features,and the gradient histogram was adopted to describe the features.Each feature was divided into specific clusters by VLAD,and data dimensionality was reduced by using local sensitive hash(LSH)to randomly project the features to low-dimensional binary signatures.Image comparison was achieved by calculating the Hamming distance on the binary image signature.Experimental results show that the proposed system can adapt to the changing environment robustly with only a small amount of memory.