Vehicle Pedestrian Detection Method Based On Data Fusion And Deep Learning

The multi-object detection technology involved in the road traffic environment mainly detects vehicles and pedestrians in images or videos collected in various road traffic environments,and implements efficient detection and recognition by means of machine learning and deep learning.This technology is a hot issue used in theoretical research and practice in recent years,and is widely used in the field of vehicle driving assistance systems,unmanned driving and the like.However,the road traffic environment is very complicated.Obstacles such as vehicles and pedestrians that need to be detected are easily interfered by various random factors,and research on multi-object detection technology still has a long way to go.In recent years,deep learning has achieved unprecedented development,so the detection algorithm of vehicle pedestrians based on deep learning in road scenes is correspondingly proposed.The parameter optimization of the algorithm is mainly related to the required training data.The specific method is to train the training data and the corresponding tags in the data set used as the input of the convolutional neural network,and then use the test in the data set after obtaining the training model.The data is obtained in the training model and compared to the label of the test data.In addition,the means of multi-sensor fusion can make up for the deficiency of single data,thus improving the detection accuracy.Therefore,this paper studies the detection of road vehicles and pedestrians based on deep learning algorithm and multi-sensor data fusion strategy.1.Based on the convolutional network residual network Resnet Net,a convolutional network that integrates multiple data is built,the data collected by multiple sensors is fused.This structure fully considers the independence and correlation of RGB image and depth image,because the structure of two convolution networks is the same and the weight is shared so that the model parameters are reduced,thus this way can improve the calculation efficiency.Fusion improved convolutional networks are trained.The experimental results show that the fusion convolutional neural network has a significant improvement in the accuracy of image classification tasks.It can lay a certain foundation for the construction of the road vehicle and pedestrian detection network.2.This paper uses the residual convolution network of multi-modal data to replace the basic network framework VGG-16 network in the original SSD network framework.Under the KITTI dataset,the road vehicles,pedestrians and bicycles are tested.After the modified SSD road object detection network model is trained and tuned,the experiment proves that the fusion method is effective.The modified fusion object detection network model is better in robustness and real-time performance,and has a certain improvement in accuracy and speed.