Research And Implementation Of Pedestrian Detection Algorithm For Zoom Scene Video Based On CNN

Pedestrian detection is the technical cornerstone of artificial intelligence with broad application prospects.With the rise of deep learning,a detection algorithm based on convolutional neural networks has emerged in the technology field of pedestrian detection.Pedestrian detection algorithms are still unable to achieve high-performance and real-time pedestrian detection at the same time,many restrictions of pedestrian size,and complex background need to be solved.For the further improvement of pedestrian detection technology,the CNN-based pedestrian detection system under the zoom scene video has certain reference value.This paper studies the pedestrian detection algorithm and implementation based on CNN in the zoom scene video.The main work and innovations are:(1)According to the problems of the complex and diverse background of pedestrians and the fact that the existing database does not fully meet the pedestrian detection requirements of the paper,the paper selected the starlight network camera to complete the zoom video capture platform construction and multi-scene video capture.The annotation software designed and implemented by Matlab is used for the pedestrian sample annotation,and the local database is constructed based on the existing pedestrian detection database.The local database construction was built based on the existing database.The experimental results show that the joint database solves the complex and variable problem of the pedestrian background to a certain extent,and is more conducive to the realization of the pedestrian detection algorithm.(2)The paper designed the SSD pedestrian detection algorithm based on CNN,and improved the network for the problems.According to the implementation of algorithm and the problem of the original SSD network which is loss of pedestrian information and impossible to detect the pedestrian of small size,and that the default bounding box of the original algorithm is redundant,a SSD pedestrian detection algorithm based on ZFNet was designed.The method of adjusting the network structure to 512×512 and increasing the pedestrian prior information reduced the loss of pedestrian information and the redundancy of the bounding box,and reduced the miss rate.According to the performance degradation problem of SSD algorithm based on ZFNet,the method of extracting lower-level feature maps and merging multiple convolutional layer output feature maps improves the detection performance of the algorithm.The experimental results show that compared with other SSD algorithls,the proposed SSD algorithm is lighter,faster,and better in detection performance.(3)According to the problem of pedestrian detection algorithm implementation,comparing and analyzing the performance parameters of the zoom camera,the selection of the Hikvision network high-defmition camera algorithm implementation platform was completed.The Pedestrian Detection Model of the algorithm training and the designed algorithm implementation flow were used to perform pedestrian detection under the real-time acquisition video of the platform.The detection results showed that the platform has completed pedestrian detection and adjustment of the zoom camera.