提出了一种结合轮廓转动惯量和特征包(BoF)算法的激光主动照明目标识别方法。介绍了转动惯量的定义，并提出了一种多尺度轮廓转动惯量特征区域检测方法和轮廓转动惯量局部不变特征提取方法。多尺度轮廓转动惯量特征区域检测方法能够提取出包含轮廓的最小特征区域，而轮廓转动惯量局部不变特征能够很好地描述轮廓的大小、位置、规则度等信息，对于各种图像变换具有不变性，并且计算效率较高。使用BoF算法统计图像的轮廓转动惯量局部不变特征，生成归一化特征直方图作为整幅图像的特征向量，输入训练好的支持向量机分类器进行识别。实验结果表明与基于Hu 矩和BP 神经网络的目标识别方法相比，所提算法在旋转和仿射变换下的识别率分别提高7.33%和19.08%。

A novel recognition method for laser active lighting system based on contour torque features and bag of features (BoF) is proposed. The concept of torque is introduced and a multi- scale contour torque feature region detector and a contour torque local invariant feature descriptor are proposed. The multi- scale contour torque feature region detector can extract the smallest feature region that contains the whole contours. The contour torque local invariant features can commendably represent the size, position, shape regularly of the contours, and they are also invariant to image transformation. What′ s more, the features are efficiently to compute. The contour torque local invariant features of an image with BoF algorithm is added up to generate normalized feature histogram, which is then input into the trained support vector machine (SVM) for recognition. The experimental results indicate that compared with existing laser active lighting recognition algorithm based on Hu moment and BP neural network, the recognition rate is increased by the proposed method by 7.33% in rotation transform and 19.08% in affine transform, respectively.