Vehicle Detection and Type Classification Based on CNN-SVM

In this paper, we propose vehicle detection and classification in a real road environment using modified improved AlexNet. Among the various challenges faced, problem of poor robustness extracting candidate regions through single feature is solved YOLO deep learning series algorithm used to potential further improve speed detection. For this, lightweight network Yolov2-tiny chosen as location network. training process, anchor box clustering performed based on ground truth set, which improves its performance specific dataset. The low accuracy after template-based extraction optimal description extracted convolution neural learning. Moreover, AlexNet, adjusting parameters, an was proposed whose model size smaller faster than original Spatial Pyramid Pooling (SPP) added solves due image distortion caused by resizing. By combining CNN with SVM normalizing features SVM, generalization ability improved. Experiments show that our method has better type classification.