Research On Pedestrian Detection In Video Surveillance Considering Perspective Rule

Pedestrian detection is the main purpose of intelligent monitoring.The traditional pedestrian detection methods are generally used to the pedestrian detection in the indoor video scenes or the smaller range of video scenes,so these methods don't consider the influences of the perspective phenomenon.With the huge increase in pedestrian trips and activities,video monitoring scenes are also gradually expanded from indoor scenes or the smaller range of scenes to outdoor scenes or the larger range of scenes.In the outdoor scenes or the larger range of video scenes,the perspective phenomenon is so obvious that it affects the performance and accuracy of the detection methods.Therefore,taking into account the effect of perspective phenomenon on the accuracy and performance of the detection method becomes extremely important.Firstly,this paper get the related geographic information through the mutual mapping methods between the video image and the geographical space.Secondly,the geospatial area constraint is introduced in the traditional methods for pedestrian detection so that the accuracy and performance of the detection methods can be improved further.The content and conclusions of this paper include:(1)The mutual mapping between the video image and the geographical spaceThis paper design experiments to compare the accuracy of the four mutual mapping methods in the situation of invariant terrain and slow change of terrain.Then,it selects the suitable method to accomplish the mutual mapping between video space and geographical space and obtains the information of geographical space area constraints.(2)Background subtraction using geographic space area constraintsFor the influences of the perspective phenomenon on the background subtraction methods,this paper uses the geographical space area constraint information to improve the methods of motion detection based on background subtraction and improve the accuracy and the performance of the methods.Then,the contrast experiments are designed to compare the accuracy and performance of the method before and after improvements from the two aspects of the intuitive side and the evaluation system side.(3)The pedestrian detection based on HOG features using geographic space area constraintsFor the influences of the perspective phenomenon on the pedestrian detection methods based on HOG features,this paper uses geographic space area constraints to improve pedestrian detection method and improve the accuracy and the performance of the method.Then,the contrast experiments are designed to compare the accuracy and performance of the method before and after improvements from the two aspects of the intuitive side and the evaluation system side.The results show that the mapping method based on the homography matrix is more suitable for video image space mapping to the geospatial space.The background subtraction method using geographic space area constraints and the pedestrian detection method based on HOG features using geographic space area constraints take into account the effect of the perspective phenomenon.As the same time,the accuracy and performance of the methods are better.