Pedestrian detection feature extraction method in road traffic auxiliary driving environment

Abstract

The invention discloses a pedestrian detection feature extraction method in a road traffic auxiliary driving environment. The pedestrian detection feature extraction method comprises the following steps of S1, establishing a positive sample library and a negative sample library used for training, and performing normalization processing; S2, calculating two layers of HOG feature vectors of each image from the sample library images; S3, combining all feature vectors of positive and negative samples into an HOG feature matrix V for the first layer HOG feature vector v calculated in the step S2; S4, performing symmetrical feature calculation on the second layer HOG feature vector w calculated in the step S2, extracting an HOG symmetrical feature vector s, and combining HOG symmetrical feature vectors of all sample images into a symmetric matrix S; S5, performing serial connection on obtained two feature matrixes V' and S, and combining into a new feature matrix Q; S6, using the feature matrix Q to train a support vector machine (SVM) classifier; and S7, adopting the SVM linear classifier to detect traffic road images. The pedestrian detection feature extraction method has the advantages of simple principle, easy realization, high detection speed, high accuracy and the like.

Description

technical field [0001] The invention mainly relates to the field of road traffic intelligent assistance, in particular to a pedestrian detection feature extraction method in a road traffic auxiliary driving environment. Background technique [0002] With the rapid development of the economy, the number of cars is increasing, which brings convenience to people's life, but also brings frequent traffic accidents. According to statistics, about 10 million people are injured in road traffic accidents around the world every year. The occurrence of these traffic accidents not only caused serious economic losses, but also brought serious hidden dangers to people's lives. In order to effectively reduce or avoid pedestrians on traffic roads being injured by vehicle collisions, domestic and foreign research institutions and automobile R&D manufacturers are paying more and more attention to the research in the field of pedestrian detection technology. For this reason, pedestrian detect...

AI Technical Summary

Problems solved by technology

At the same time, because the HOG feature dimension is too high, the detection speed is slow, which cannot meet the real-time application requirements

[0008] (3) For example, Dollar et al. used integral histogram technology to quickly calculate HOG features, proposed integral channel feature (ChnFtr), combined with AdaBoost classifier, which greatly improved the detection speed, but it took a long time to train the classifier

Although this method improves the speed of pedestrian detection, it loses part of the feature information of the image, and has a higher false detection rate than Dalal's method.

[0009] A method based on a mixture of templates and statistical classification: This method can improve the detection accuracy, but the features only increase, and the calculation of features and classifiers also increase the prediction time, and the real-time requirements cannot be met.

Images