Improvement and Comparison of Traditional CNN and SVM Classification Based on Hog Descriptor in Pedestrian Detection
Yusen Luo, Xin Liu, Xuhang Cao · 2021
This paper first briefly introduces and compares several traditional convolutional neural networks. Next, it presents the Hog (Histogram of oriented gradient) Descriptor. Then, it mainly discusses and compares the differences between HOG + SVM algorithm and traditional convolutional neural network in the field of pedestrian detection. And finally, it describes how to improve the traditional HOG + SVM with Non- Maximum Supp and bootstrap method. Experiments show that the accuracy of the improved HOG + SVM algorithm is much higher than that of the traditional CNN, and the time required is not much