基于Bag-of-Features算法的车辆检测研究

A study of vehicle detection based on Bag-of-Features algorithm

车辆检测已成为交通运输工程(ACC)和先进辅助驾驶系统(ADAS)中的核心技术之一。该算法利用车辆的边缘特征与Bag-of-Features(BoF)模型的融合对前方运动车辆进行实时检测,主要包含车辆假设存在区域生成和假设区域验证两部分。首先,对图像进行预处理后利用Sobel边缘检测处理得到车辆假设存在的区域;然后,利用Bag-of-Features的K最近邻域算法对假设存在区域进行验证。该算法与其他算法最大的区别在于将边缘和Bag-of-Features相结合来提高检测率。通过对实际道路视频进行测试,结果表明,该方法能够实时准确地检测出道路上前方运动车辆。

Vehicle detection has become one of the core technologies of Automatic Cruise Control （ACC） and Advanced Driver Assis- tance Systems （ADAS）. This algorithm makes real-time detection for the front moving vehicles with the combination of vehicle edge features and Bag-of-Features （BoF） model, which mainly includes two parts-generating the assumed region in which vehicle exists and verifying the assumed region. Firstly, it deduces the assumed region of vehicle with Sobel edge detection processing after the pretreat- ment of image. Then, by using the K-nearest neighbor algorithm of Bag-of-Features verifies the assumed region of vehicle. The differ- ence between this algorithm and other algorithms is that this algorithm adopts both edge features and Bag-of-Features, which can effec- tively improve the detection rate. The test result from the road-condition videos has shown that the method can make real-time detec- tion of the front moving vehicles on the road accurately.