Research On Machine Learning Method For Accurate And Efficient Location Of Region Proposal In Object Detection

The object proposal,as the basis of object detection and object recognition,detect the initial positioning of the object,and is suitable for positioning multiple types of object as well as the single-type object.Object proposal plays an important role in the actual environment,such as real-time monitoring systems,access control systems,brush face payment,etc.Therefore,the accurate and efficient generation of the object proposal has an important research significance.This paper mainly studies the efficiency and accuracy of object proposal generation.(1)This paper mainly focuses on the low accuracy of BING method,Firstly,extracting feature by using the extended template to enhance the discriminate of the edge features of the obj ect and improve the bias problem of the BING.Then the Gaussian soft weight non-maximum suppression algorithm is used to solve the problem that the non-maximum object candidate region near the object proposal is forced to zero,and the accuracy of the BING method is improved.Finally,the improved algorithm is verified on the public data set.The results show the improved BING improves the accuracy of the BING-based object proposal generation without increasing the computational of the algorithm..(2)This paper improves on two aspects based on the selective search method with higher accuracy.The One is to use the Hamming distance to calculate the similarity between adjacent pixels in the initial segmentation stage,and the other is to select the merge strategy.To solve complex problems,a color-based hashing strategy is used to calculate the similarity of the initial segmentation regions of the image to determine whether adjacent regions can be merged.The approach greatly reduces the number of iterations that the selective search algorithm computes during region merging and validates it on the exposed test set.The results show that the improved selective search method improves the efficiency by 60%on the basis of not reducing the accuracy as much as possible.Based on the application of the obj ect proposal in the DPM pedestrian detection algorithm.In this paper,the improved BING algorithm and the improved selective search algorithm are used to replace the sliding window in the DPM pedestrian detection algorithm to complete the detection of pedestrians.The test results are verified by the pedestrian INRIA dataset.The results show that the improved object proposal combined with DPM method is used to detect pedestrians,which speeds up the DPM pedestrian detection and achieves efficient and accurate positioning of pedestrians.