Detecting Vehicles And Pedestrians By Using Single Shot Multi-box Detector

For the last few years,auto-driving technology has been greatly developed and becoming one of the hottest research field all around the world,what's more,object-detecting of the road traffic situations is getting more and more attention as one of the key tasks of auto-driving technology.By now,there are several ways used for object detection,thanks to these solutions,drivers can get much more precise traffic information,which not only can greatly avoid the happening of potential traffic accidents but also can provide powerful protection on people's life and property security.The main task of object-detecting for auto-driving technology is to detect the vehicles and pedestrians on the traffic environment.Unfortunately,sometimes,vehicle and pedestrian objects become distortion because of the environmental factors such as distortion,covering,attitude and etc,which bringing much more challenge for detecting the vehicles and pedestrians on the traffic environment.The traditional method for object-detecting is to compare the image of the environment captured by the image acquisition equipment with the limited sample images in the certain database,the detecting result can be easily affected by the external factors and more importantly,the detecting speed and detecting accuracy can hardly be improved,so the traditional method is not suit for auto-drive field.An alternative new method is based on the Artificial Neural Network(ANN).Object-detecting method based on ANN can "teach" the machine to acquire the concept of the object's feature by extracting the feature from the sample images,which can greatly improve the detecting speed and detecting accuracy,more importantly,the detecting result will not be affected by the external factors.In this paper,a implementation method is put forward after comparing the advantages and disadvantages of both traditional object-detecting method and method based on ANN.For CNN,it has obvious advantage in extracting the objects' feature;for deep learning networking model,it can get good and precise feature information of the objects;for SSD algorithm,it is one of the most popular algorithms in the object detecting field,because of its fast detecting speed and accuracy.For there is no public dataset about the vehicles and pedestrians on the traffic environment,in the experiment part,the dataset is collected and collated by the author one by one.With this same dataset,the detecting results of YOLO algorithm and SSD algorithm are compared which proves the advantage of SSD algorithm in the field of object-detecting.At last,based on the deficiencies of the experiment,the improvement direction is provided.