Research And Application Of Vehicle Surface Scratch Recognition Technology

As the demand for imported automobile market in China has sharply increased,it is inevitable that scratches will occur during the transportation process.Therefore,complex surface scratches must be identified before entering the carrier warehouse.At present,the recognition of automobile surface scratches still adopts the manual visual measurement method,which has the disadvantages of low efficiency,high work intensity and high leakage rate.Therefore,computer vision detection technology is urgently needed to assist or replace manual detection.In this article,the system is designed to assess the damage of the vehicle surface scratch based on image processing and artificial intelligence,and we focus on the key technologies of vehicle surface image de-dusting,car surface scratch recognition and scratch classification and so on.It can meet the performance requirement of the system in terms of efficiency,accuracy and reliability.The main research contents of this thesis are as follows:(1)The requirement of the vehicle surface scratch insurance estimation system is analyzed.The acquisition equipment,the overall framework and the main functional structure are given,and the key technologies used in the system are introduced in detail.(2)In order to tackle the problem of color distortion and contrast reduction of automobile surface image caused by atmospheric dust,in this thesis,a dust removal algorithm based on dark channel prior is proposed.Firstly,we construct a physical degradation model according to the transmission way of light in the dust layer and atmosphere.Then the white car image is inverted to meet the dark channel principle,the two model parameters included the atmospheric light and coarse transmission map are then estimated by the degradation model and dark channel prior.Meanwhile,guided filtering was used to optimize the transmission rate.Finally,on the basis of the physical degradation model,the dust of the automobile surface image is removed quickly and the image of the white automobile dust removal is reversed.The experimental results show that the algorithm is applicable to dust removal of automobile surface image.The color and contrast of automobile surface image can be recovered by the reconstructed image.(3)Aiming at the disadvantages of discontinuous edge detection and less detection information of traditional Sobel operators,an edge detection method based on improved Sobel operators was designed.Improved algorithm extends the original two orientation templates to four orientation templates,the45~o and135~o direction characteristics are added to improve the continuity and accuracy of automobile edge detection.Experiments show that this method meets the requirements of accuracy and high efficiency of automobile edge detection.(4)Aiming at the problem that the classification of automobile surface scratches based on BP neural network is easy to fall into local minimum value and the number of hidden layer nodes is difficult to determine,an algorithm for car scratch classification based on optimized BP neural network is proposed.The number of hidden layer nodes is determined by the golden section algorithm.Then the weight and threshold of BP neural network are optimized by the improved genetic algorithm according to the advantages of genetic algorithm in global optimization.The optimized BP neural network is used to judge the scratches on the surface of the car.The example analysis shows that the algorithm can accurately and effectively detect the type of scratches on the surface of the car and improve the fault tolerance of the scratch classification.(5)The main functions of the vehicle surface scratch damage determination system is realized.The key technology realization and important operation interface of the main function modules of the system are given.