改进的卷积神经网络行人检测方法

Improved convolutional neural network pedestrian detection method

提出基于YOLOV3和DenseNet相结合的轻量化行人检测算法。加入HSV图像处理模块强化行人特征,利用卷积神经网络提取行人特征,通过k均值聚类算法筛选预测框,借鉴特征金字塔的思想做高低层特征融合和预测,利用Dense Block结构对网络轻量化进行完善,在国际广泛使用的行人数据集上进行一系列实验。实验结果表明,检测速度比现有的优秀目标检测模型YOLOV3提升了8倍,模型大小为YOLOV3的1/107,所提方法在测试集上的实时性和准确率都有所提高。

lightweight pedestrian detection algorithm based on YOLOV3 and DenseNet was proposed.The HSV image processing module was added to enhance pedestrian characteristics,and the convolutional neural network was used to extract pedestrian characteristics,k-means clustering algorithm was used to screen out the prediction box,the features of pyramid networks were brought up for the feature fusion and prediction of high and low layers,the Dense Block structure was used to improve the network lightweighting,and a series of experiments were conducted on the widely used pedestrian dataset.Experimental results show that the detection speed is 8 times higher than the existing excellent target detection model YOLOV3,and the model size is 1/107 of YOLOV3.The proposed method improves the real-time performance and accuracy of the test set.