Research On Pedestrian Detection Based On Improved Mr Significant Information

With the advent of the Internet big data era,video surveillance technology has been slowly spread around the world.The technology can serve many areas to facilitate people's lives and work.Pedestrian detection technology is a key technology in the field of intelligent monitoring,which can be used in intelligent assisted driving,robot visual orientation,etc.,which makes pedestrian detection have great commercial development value.Up to now,pedestrian detection algorithms have not been fully applied to pedestrian actual products.The reasons for this can be summed up in two aspects:accuracy and real-time,that is,the accuracy can not be guaranteed and the requirements of online real-time can not be achieved.Researchers have been working to overcome these two issues.This paper summarized and studied the history of the development of pedestrian detection algorithms,and uses the HOG features with good stability and robustness to conduct pedestrian detection algorithms.In view of the above two major problems,respectively,from the perspective of significant extraction and improvement of HOG features to solve a proposed based on improved MR significant pedestrian detection algorithm.The main research work in the thesis has the following aspects.?1?The local HOG features are improved,and the PHOG features are proposed.This feature is more capable of describing the layout target and the background information,making the difference between the background and the target greater,and the classifier more easily identifying pedestrians.PHOG features can increase the recognition rate of the window target and reduce the missed detection rate of pedestrian detection.In addition,the histogram equalization is used in the preprocessing,which can effectively adjust the contrast of bad images and make the profile information of pedestrians more exposed,which is beneficial to the result of pedestrian detection based on PHOG features.?2?Based on the PHOG features,incorporating the Objecteness features and PCA dimensionality reduction,the O-PHOG-PCA features are proposed.The Objecteness feature can quickly eliminate non-object targets,and the feature dimension is low,and the complexity of the features after integration is small.The O-PHOG-PCA feature dimension is short,which reduces the time of feature recognition,thus reducing the complexity of the overall detection time.The method of establishing characteristic pyramid is proposed,which makes pedestrian detection unnecessary to extract multi-layer features and reduces the time of feature extraction.?3?The traditional MR saliency algorithm is improved,the screening method of real background seed points is proposed,and the significant target is perceived based on the real background seed points,which improves the detection rate and recall rate of the target area.According to the improved detection method of MR significance,the area where pedestrians may exist was extracted.The pedestrian detection in this area is prioritized,which greatly reduces the detection range and reduces the running time of the overall pedestrian detection algorithm.Finally,we verified the effectiveness of our algorithm through a variety of pedestrian detection experiments.The experimental results show that the algorithm reduces the missing detection rate from 35%to 22%at FPPI of 10^-2.At detection speed,the pedestrian detection algorithm achieves 112 ms/frame at 350*280 resolution,which is better than the traditional feature-based pedestrian detection algorithm.The experimental results show that the proposed algorithm can effectively reduce the missed detection rate and has the advantage of detection speed compared with the traditional algorithm.