面向车辆识别的样本自反馈式级联检测方法

以解决机器学习的样本提取问题、提高车辆识别的精度为目的,以Adaboost算法、级联检测方法、样本自反馈方法为基础,提出了一种面向车辆识别的样本自反馈式级联检测方法。方法兼具Adaboost级联检测器的架构优势和样本自反馈方法的样本扩充能力。通过前车图像的识别实例阐述了样本自反馈式级联检测方法的识别过程,进而使用常规Adaboost检测器和样本自反馈式级联检测方法对图像数据集进行了识别,将生产者精度、用户精度、总体识别精度、k值作为评价指标对识别效果进行了比较,证明了样本自反馈式级联检测方法具有较高的识别精度。分析了样本自反馈式级联检测方法个别评价指标较低的原因,提出了相应问题的解决思路。

Feedback Samples Based Cascade Method for Vehicle Detection

The principle of Adaboost algorithm,cascade method and sample self feedback method are analyzed and used to build a feedback samples based cascade method that can detect vehicles without problems of sample extraction and have high accuracy. The method has the advantage of the Adaboost cascade detector and the sample expansion ability of the sample self feedback method. The detection process of the self feedback cascade detection method is illustrated by the example of front vehicle image detection. Image data sets are detected by conventional Adaboost detector and feedback samples based cascade method,and the Producer's Accuracy,the User's Accuracy,the Overall Accuracy and k are used as evaluating indicators to compare test results,and the results show that the feedback samples based cas-cade method has higher accuracy. The reason of low evaluation index of feedback samples based cascade method is ana-lyzed,and corresponding solutions are proposed.