Research on the Cascade Pedestrian Detection Model Based on LDCF and CNN
Zhonggui Ma, Pan-pan Gao · 2018
In recent years, with the rapid development of deep learning, pedestrian detection have brought great vitality. Convolution Neural Network (CNN) can be used for pedestrian detection to improve the detection performance, but the CNN model needs a lot of parameters and has high computational complexity. In order to settle these problems, the cascade pedestrian detection model based on Local Decorrelation Channel Features (LDCF) and CNN is proposed in this paper. In the first stage, the LDCF is used to obtain the pedestrian candidate window, and it can reduce the redundancy of the image, decrease the computation complexion of CNN. In the second stage, the output results of LDCF as the input of the CNN model, then CNN model is trained and an LDCF+CNN cascade detector is obtained by cascading the LDCF detector with the CNN model. The experimental results show that the log-average miss rate of LDCF+CNN algorithm is lower than that of LDCF algorithm alone, and the miss rate is below 13.21%.