Research On Character Cascade Classifier Based Adaptive Detection Algorithm For Automobile Dashboard

Computer vision inspection has been very common in industrial production.With the continuous development of computer vision technology,the computer vision detection equipments have been developed in the direction of precision,generalization,and automation.Establishing a vision inspection system with high generalization ability and reliability is the key to improving the production efficiency and reducing the production cost.This research comes from the problems encountered by automobile dashboard manufacturers using traditional automobile dashboard detection algorithms.Traditional automobile dashboard detection algorithms need to change a large number of parameters before they can be used to detect different types of automobile dashboard.This kind of design is very time-consuming and the algorithm is not universal,obviously cannot meet the production needs of the dashboard manufacturer.In addition,traditional automobile dashboard detection algorithms have low detection accuracy and algorithms are not robust to image noises.Therefore,establishing a dashboard vision inspection system with high precision,high universe,and high reliability is an urgent need for dashboard manufacturer.This paper aims at enhancing the detection accuracy and generalization ability of the automobile dashboard detection algorithm.An adaptive detection algorithm based on the character cascade classifier is proposed.This paper proposes a brand-new automobile dashboard detection framework and this algorithm can automatically complete the detection process of different automobile dashboards.This paper first analyzes why pointer extraction algorithm is easily affected by the number on the dashboard and then a contour analysis based algorithm is proposed to precisely extract pointer.After that,a line-fitting algorithm based on least square method is used to estimate position of the pointer,which improves the accuracy of the pointer extraction.For the traditional automobile dashboard detection algorithms are not universal,these algorithms need to modify a large number of parameters to detect different types of automobile dashboard.This paper proposes an adaptive character cascade classifier based detection framework.By identifying the speed characters corresponding to the main tick marks in the speedometer and tachometer first,and then finding the tick marks on the dashboard based on character.This paper proposes a tick marks extraction method based on region search strategy.This new design greatly improves the generalization ability of the automobile dashboard detection algorithm.The experiment shows that the character recognition result of the traditional HOG(Histogram of Gradients)-SVM(Support Vector Machine)classifier is usually poor.The concept of character cascade classifier is proposed and the character filter is designed.The HOG-SVM classifier,character filter,and CNN(Convolutional Neural Networks)character classifier are cascaded to identify digit characters and this design can significantly improve the character recognition accuracy of the dashboard detection algorithm.Due to the parallax error,the reading error using previous algorithm is too large.The linear description relation based on the Newton interpolation is introduced to describe the relationship between the pointer position and its reading value.This design rule is simple and clear,and the detection error of the algorithm is lower than most algorithms and this algorithm is robust.The experimental results show that the algorithm can significantly improve the detection accuracy.Compared with other dashboard detection algorithms,the generalization ability of this paper is greatly enhanced.This detection algorithm can be used to detect different types of dashboard without changing a large number of parameters.The adaptive detection algorithm proposed in this paper has broad industrial application prospects and can significantly improve the manufacturer's production efficiency.