Obstacle Detection Method Based On Multi-source Image Fusion

With the development of technology,intelligent transportation has become the main trend of transportation industry in the future.Obstacles detection is an important technology for intelligent driving.It mainly detects obstacles such as pedestrians and vehicles in the field of view that affect driving safety.At present,intelligent safety driving system is generally based on multi-sensors,such as visible image sensor,microwave radar and laser radar etc.However,the visible image sensor cannot work in the all-weather complex environment,and the laser radar is greatly affected by rain and snow.In order to better assist the driver in driving safely and realize intelligent obstacles detection,this thesis studies an obstacle detection method based on the fusion of visible image and infrared image.The complementary information of visible image and infrared image are utilized comprehensively to improve detection performance of obstacle detection system.And it makes detection system better adapt to different road conditions.The prerequisite of image fusion is image registration.In term of image registration,compared with visible images,local contrast inversion exist in infrared images,which affect the accuracy of image registration.To solve this problem,this paper adopts descriptor recombination strategy to solve the problem of image local contrast inversion.The Singular Value Decomposition(SVD)is used to reduce the dimension of the feature descriptor of feature points detected by the SIFT algorithm to improve the matching efficiency of the algorithm.At the same time,an improved random sampling consensus algorithm is used to eliminate mismatched point pairs,which improves the matching accuracy of matching points.In the aspect of image fusion,the Non-Subsampled Shearlet Transform(NSST)is used to decompose the input image into high-frequency sub-band information and low-frequency sub-band information.In order to keep more texture information in the fused image,an improved Pulse Coupled Neural Networks(PCNN)was used to fuse the high-frequency subband images.For the low frequency subband images,the Hypercomplex Fourier Transform(HFT)visual saliency model is used to extract the salient regions.The image similarity and image information entropy ratio are used as the fusion basis,and the salient regions and non-salient regions of the low frequency subband images are fused respectively.Finally,NSST inverse transformation is used to obtain the fused image.In this paper,the method based on gradient statistics and the Multiple Visual Salient Fusion(MVSF)method are used to locate the obstacles in the detection areas firstly.Then the dual-threshold Canny edge detection operator is used to eliminate interference contours.Finally,the sparse optical flow algorithm(Lucas-Kanade,LK)is used to track and accurately locate the obstacles in the warning area and the detection areas.The experimental results have shown that the algorithm proposed in this paper can effectively detect and locate the obstacles in front of and on the sides of the vehicle.And it has a high detection accuracy rate.