Research And Implementation Of Pedestrian Detection

Pedestrian detection is a technique for determining whether a pedestrian appears in an input image or video sequence and determines its location.It is a challenging research hotspot in the field of computer vision,and now plays an important role in intelligent assisted driving,pedestrian analysis and intelligent video surveillance.However,because of the different shapes of human body,the diversity of dresses,and the frequent changes in lighting,climate change,and occlusion of scenes,the study of pedestrian detection has become a hot topic.The most widely used pedestrian detection method is based on the combination of Histogram of oriented gradient(HOG)and Support Vector Machine(SVM).Because the HOG feature dimension is too high,it affects the detection speed.Support vector machine is difficult to meet the needs of daily production in terms of detection rate.Aiming at the increasing requirements for the accuracy and detection efficiency of pedestrian detection.On the basis of previous research,a pedestrian detection method based on HOG+SVM is proposed.Mainly optimized in the following aspects:(1)In terms of feature extraction,in view of the superiority of HOG feature in texture,we choose to extract the HOG feature of the sample.Because the HOG feature dimension is higher and the speed is slow,we use PCA algorithm as far as possible to reduce the dimension and improve the detection speed in real time.(2)SVM is used as a classifier for classification,SVM based on traditional single-core function is optimized,SVM of combined kernel function is adopted,and penalty factor and slack variable are introduced to achieve better classification effect.(3)Using the improved genetic algorithm and particle swarm optimization algorithm to optimize the parameters of the support vector machine,and compare the other optimization methods to find the most suitable method.(4)Using the Adaboost algorithm,the classification effect of the parameter-optimized support vector machine is further improved.The support vector machine with the optimal parameters is used as a weak classifier,and finally becomes a strong classifier through layer-by-layer iteration.This Adaboost is used as the final classifier.In this paper,after the improvement of feature extraction and classifier training,pedestrian detection experiments are carried out under datasets such as IN-RIA database and MIT database.It can be seen from the test results that in the simple background,the proposed algorithm is much better than the traditional algorithm for occlusion problems and pedestrian detection in different poses.Under complex background,the algorithm is for pedestrian detection in rainy,snowy,foggy and other environments.The effect is also much better than the traditional algorithm.The final data shows that the detection rate of the final algorithm is 97.3%,which is 5.8% higher than that of the traditional algorithm,and at least 1.7% higher than other optimization algorithms,and the detection speed reaches 0.23 s,which meets the requirements of real-time processing.