Research On Video-based Pedestrian Target Detection And Recognition

Pedestrian detection and recognition is an important research field in computer vision,which is of great significance to the improvement of human living standard.It plays an important role in the fields of automatic driving,intelligent monitoring,human-computer interaction and motion analysis.The difficulty of pedestrian detection is different from other target detection because of the complex background,changing illumination,uncertain pedestrian posture and diversified clothing in actual scene.Based on the analysis of the existing algorithm under the premise of applying computer vision and digital image processing knowledge,puts forward the improved algorithm of image pedestrian detection and recognition is an important research field in computer vision,and to the improvement of human life today has the great significance,in automatic driving,areas such as smart surveillance,human computer interaction and motion analysis play an important role.The difficulty of pedestrian detection is different from other target detection because of the complex background,changing illumination,uncertain pedestrian posture and diversified clothing in actual scene.Based on the analysis of the existing algorithm under the premise of applying computer vision and digital image processing knowledge,puts forward the improved algorithm to detect pedestrians targets in the images and video,finally,the moving pedestrians in video tracking,in this paper,the main work is as follows: and pedestrians target detection in video,finally,the moving pedestrians in video tracking,in this paper,the main work is as follows:In this paper,a mixed Gaussian background model based on HSV space is proposed.The image is transformed into HSV space,and the value of brightness(V)component in the background model is compared with the threshold value,so as to achieve the effect of suppressing the shadow of pedestrian target.By comparing the experimental results of three consecutive frames in the same video with different algorithms,the accuracy of the method in this paper is verified,which provides a good foundation for the follow-up research.Then,the integral graph is used to simplify the HOG feature extraction,which avoids the repeated calculation of the pixel gradient information of the overlapping parts between blocks.In this paper,bilinear interpolation is used to determine the gradient direction and the amplitude of the corresponding direction,which can degenerate the image details and better locate the pedestrian as a whole.For the problems of traditional HOG feature detection speed is slow and the extracted feature vector dimension is high,the principal component analysis(PCA)algorithm is used to reduce the dimension of HOG to avoid the influence of redundant information of feature vector on the classification process.The improved algorithm can effectively improve the recognition rate and reduce the detection time.Finally,the Mean Shift algorithm is used several times in the video frame sequence to adjust the position and size of the rectangular frame according to the position and shape change of pedestrians.When the pedestrian target is tracked,the Kalman filter only detects the motion information of the moving human target,while the improved Mean Shift algorithm detects the color features of the moving human target.In this paper,we propose to combine the two methods and use Kalman as the improved Meanshift algorithm to predict the position,and use the improved Meanshift as the observed value to continuously correct the coefficients of the Kalman equation.