Research On Pedestrian Detection Problem Based On Clustering Algorithm

With the development and progress of computer technology,people need to process more and more images,and the application of images is becoming more and more extensive.Therefore,image processing has gradually become a popular direction in computer vision.In image processing,the goal Testing is extremely important.In target detection,pedestrian detection is a special case of detection.In complex scenes,the detection of a single pedestrian is still a challenging problem,because pedestrians will gather together,so occlusions often occur.Happening.In order to solve this problem,this article has studied the Faster R-CNN algorithm in depth,and made relevant improvements based on Faster R-CNN.This article mainly contains the following two parts.First,in the pre-processing stage,a simple linear iterative clustering algorithm branch is proposed,and the algorithm is used to improve the speed and accuracy of pedestrian detection.Secondly,in the post-processing stage,a constraint condition is added to the non-maximum value suppression algorithm,and the generated multiple candidate frames are non-maximum value suppression to further improve the detection performance.(1)Pedestrian detection based on clustering algorithm.Due to the complexity and variability of the scene,the difficulty of pedestrian detection is gradually increasing.In order to better detect pedestrians,first in the input picture of Faster R-CNN,each pixel in the picture is divided into super pixels,and then the super pixels The segmentation result and RGB stitching are used as the input of Faster R-CNN for subsequent pedestrian detection.The experimental results show the real effectiveness of Faster R-CNN pedestrian detection research based on the clustering algorithm.The results show that the algorithm can effectively exceed the benchmark and accurately locate and classify the pedestrians to be detected.(2)Improvement of non-maximum suppression.In the non-maximum suppression part of the post-processing stage of Faster R-CNN,the traditional improvement is also carried out.Because the traditional non-maximum suppression removes all the detection frames whose intersection ratio is higher than the threshold,in dense pedestrians,This approach is very likely to cause missed detection,so this article changes the original overlap threshold calculation method,adds a new overlap threshold and a detection frame score threshold,and proposes a new constraint method with one line of code,which reduces the traditional non-maximum value.Suppress the missed detection and false detection of the target caused by it.After a large number of experiments on two real data sets,the results show that the Simple-NMS proposed in this paper is better than the comparison algorithm in the detection results.