Pedestrian Detection By Convolutional Neural Network Using Continuous Frames

Pedestrian detection is an important research topic in computer vision.It is widely used in many fields,such as public security,scene monitoring,traffic operation and so on.Most of the traditional pedestrian detection methods use artificial features to describe human characteristics,such as gradient,texture,edge and so on.On the basis of separating the foreground objects from the background in the image,a pre-built human body model is used to detect pedestrian on the image.However,these methods are usually too complex,and are easily affected by the environmental factors and the changes of human form.In particular,artificial features are difficult to adapt to complex and dynamic scenes.In practical application,pedestrian detection effect is not ideal.In recent years,the theory of deep learning has provided a new idea for many computer vision problems.Especially,deep convolutional neural network has shown excellent performance in many classification applications.Some researchers begin to pay attention to the application of deep convolutional neural network in pedestrian detection.Compared with the features designed by human,the deep neural network with data training can express the characteristics of the human body butter,so as to improve the accuracy of pedestrian detection.The problem of target occlusion in pedestrian detection has been an important factor affecting the accuracy of pedestrian detection.Researchers have proposed many methods to deal with the problem of pedestrian occlusion,but the effect is not very satisfactory.Aiming at the problem of occlusion handling in pedestrian detection,this paper proposes a method of pedestrian detection based on convolutional neural network with continuous frames.In our method,the temporal and spatial features of the video sequences are extracted using the continuous frames convolutional neural network,which is used to improve the ability of occluded pedestrian detection.At the same time,considering the detection result in time sequence,this paper will introduce the idea of tracking method for pedestrian detection.By using the tracking algorithm to generate the prediction of pedestrian position,the pedestrian detection results can be re-evaluated.Thus the error of pedestrian detection is reduced.In order to test the method proposed in this paper,pedestrian detection experiments were carried out on the Caltech Pedestrian Detection Benchmark and the Visual Tracker Benchmark.The experimental results show that our method can deal with the pedestrian occlusion problem better than the single image pedestrian detection method,which is helpful to improve the accuracy of pedestrian detection.