Technology And Realization Of High Density Pedestrian Detection Under Complex Illumination Conditions

In recent years,pedestrian detection technology has been applied in many real-life scenarios to help urban life become more convenient.With the continuous development of smart cities,pedestrian detection technology will be used in many human-computer interaction scenarios in the future.Pedestrian detection has very considerable application prospects.Hence the research on pedestrian detection is very necessary.Nowadays,there are many excellent pedestrian detection algorithms,but pedestrian detection has different difficulties in different places.For example,detection in dimly lit places is more difficult,so pedestrian detection technology also needs to be improved.Pedestrian detection technology has also developed from the traditional mode of extracting pedestrian features + classifiers to the deep learning stage.This is also a big step.Pedestrian detection algorithms based on deep learning are mainly divided into single-step detection and two-step detection.The candidate frame is separated from the classification,and the latter obtains the final detection result directly from the picture.Compared with the traditional detection stage,it has obtained a great improvement in speed and accuracy.At present,there are many practical applications of related pedestrian detection algorithms in life.The essence of pedestrian detection is to identify pedestrians in the image and frame the accurate location.Due to the complexity and change of the category of pedestrians,there are many difficulties in accurately detecting pedestrians.The existing pedestrian detection technology cannot solve all the problems.This paper mainly focuses on the following two issues to study related algorithms: one is that pedestrian density is often high when pedestrian detection in public places,and the other is that complex lighting conditions have an impact on the accuracy of the algorithm.The YOLOv3 algorithm is currently the most widely used in the YOLO series of algorithms.YOLOv3 can maintain high detection accuracy,and is faster,and can even meet the requirements of real-time.Therefore,this paper uses the YOLOv3 algorithm as the basis to achieve pedestrian detection.Around the two points mentioned,the main contents of this paper are:(1)Aiming at the problem that the characteristics of pedestrians in complex lighting conditions are not obvious,the image preprocessing technology is introduced before the detection to solve the detection difficulty caused by complex lighting.After research and comparison of several popular image preprocessing algorithms,select comprehensive The best performing MSRCP algorithm.(2)Aiming at the problem of high-density pedestrian detection accuracy,by analyzing the pedestrian detection algorithm,using the YOLOv3 algorithm as the basis,the network structure of the YOLOv3 algorithm is further modified,the fourth scale is introduced into the basic structure of YOLOv3,and the high-density pedestrian data Set Wider Person to perform dimensional re-clustering to obtain a priori frame size suitable for high-density pedestrian scale features,which further improves the detection accuracy.After training and tuning,the pedestrian detection accuracy of the algorithm model of this paper is higher,which shows that the method of this paper is effective.(3)Apply image preprocessing technology and improved pedestrian detection algorithm to develop a pedestrian intelligent detection system to realize the visualization of pedestrian detection results.The system covers the three major functions of user basic information management,image detection,and real-time video detection.The algorithm has been applied in practice.