A novel method of front vehicle recognition

Smart car shows great potential in our future life and has attracted lots of interests from many research and industry communities. In this field, the technique of machine vision and recognition plays an important role, for instance, the automatic front vehicle recognition can provide driving safety information for the smart car. Lots of previous work on front vehicle recognition has been done, most of them are based on the techniques of template matching and machine learning. In this paper, we propose a novel method of front vehicle recognition, in which the context information of the images and the machine learning method are integrated to improve the performance of recognition. In this work, the data is collected from a video camera mounted on our testbed car. With the video stream, we extract the key-frames and convert them into gray images. The context information of the images is explored by the analysis of the lane marks, in particular the K-means clustering algorithm is applied for lane marks segmentation. With the lane marks, we can drop off some non-promising areas of the image for car recognition. This can speed up our work and reduce false alarms. Then, we employ morphological processing method to locate the candidate image regions of the front vehicle. Finally, the histogram of gradient (HOG) feature is adopted in the support vector machine (SVM) for the front car classification. Experiments are conducted on a real world dataset. The experimental results show that the proposed method achieves significant improvement compared with the previous work and the baselines.