Pedestrian Detection And Counting Based On Deep Learning

In recent years,with the development of big data and computer hardware,the rapid development of deep learning has led artificial intelligence and computer vision into a golden period of development.In the field of computer vision,pedestrian detection is an important branch of target detection.Pedestrian detection Pedestrian detection has a wide range of application scenarios in many artificial intelligence fields,such as: autonomous driving,human-computer interaction,and intelligent monitoring.Compared with the traditional pedestrian detection,the deep learning method has certain advantages in the accuracy of pedestrian detection by using convolutional neural network to learn features,but there are still some problems,such as miss small-scale pedestrian detection and pedestrian occlusion problems.These problems make it difficult for pedestrian detection counting to be widely used in realistic scenarios.In view of the above issues,this paper conducts research on pedestrian detection and counting based on deep learning.The main work and innovations of this article include:(1)Establishment of datasetAt present,most datasets are collected in Europe and the United States,and the actual traffic conditions and personnel characteristics in China are different from those in Europe and the United States.In order to improve the adaptability of the actual application of the algorithm,a pedestrian detection dataset was established based on the video surveillance camera data of the skynet surveillance camera system.(2)Detection of small-scale pedestrian.The height of the monitoring equipment in the skynet surveillance camera system or the distance to pedestrians will cause pedestrians with smaller sizes to appear in the camera screen frequently.In addition,there is a situation where the resolution is not uniform,so the training picture needs to be cropped and stretched.The picture will be distorted after being cropped or stretched,which can due to the miss of small pedestrians.This paper studies pedestrian detection networks based on spatial pyramid pooling and multi-scale detection strategy.(3)Detection of occlusion pedestrian.In real life,due to the complex background and different postures of pedestrians,pedestrian occlusion problems can occur,including people being blocked by objects and people being blocked by people.As the depth of the network increases,the feature information of the blocked pedestrians will be lost layer by layer during the feature extraction and transmission of the multi-layer network.To solve this problem,this paper proposes to use Res2 Net to represent multi-scale with a finer granularity characteristics,and improve the range of receptive field in each layer.The experimental results show that the entire network can effectively reduce information loss,realize multiple-layer feature reuse and fusion of the network,and can quickly and effectively detect pedestrians with occlusion.