The Research And Development Of The Object Detection System Based On Deep Learning

The rapid development of computer science has brought great progress to human life and makes it more and more intelligent.Artificial intelligenc e has always been an important area for human exploration.As we all know,human vision is an important part of the perception of the outside world.Scientific research shows that nearly 80 percent of the information is perceived by vision.Thus,computer vision has always been an important research direction in the long way of human exploration of artificial intelligence.Computer vision involves image processing,machine learning,pattern recognition and other disciplines.The ultimate goal is to simulate the human visual,in order to use the computer to complete a variety of identification tasks.Target detection is mainly to detect the target in the picture.For example,the target detection requirements of the automatic driving system is to detect the pedestrians,vehicles and other objects in current driving environment.Due to the complexity of the real road conditions,the detection system needs a higher level of semantic understanding for the scene.In the past,most of the target detection algorithms were based on traditional filtering methods.People extracted classical features such as SIFT [22],HOG [2],and then put them into classical classifiers(such as SVM [30],Adaboost [29])for classification and identification.As the use of manual characteristics,so poor robust,and the workload is large.The target detection effect is very different when the environment changes significantly.Because of the strong ability of expressive neural network in deep learning,the extracted features have very strong robustness.Therefore,this paper mainly uses Faster R-CNN [5] which is the more classical detection framework based on deep learning.On this basis,this paper tries to use different feature extraction layer and changes the network structure based on traditional classical model.So that the current network model can make a better trade-off between the precision and speed.With using the calibration data to train the model,adjust the parameters,a better accuracy and speed of the model can be trained and applied to the detection system.The development environment of the target detection system is Linux,and uses Qt as the development framework of the interface which is focused on image interface,and C++ as the bottom.The target detection system development process described in this paper mainly includes the overall demand analysis,the overall design,implementation and testing.Finally,through the test,the system have a good performance in the hardware and performance.