Research On Method Of Pedestrians Counting In Video Surveillance

In some video surveillance application locations, such as airport, bus station, wharf, bank, hazardous area, work area, etc., for the sake of security assurance as well as work attendance recording in work and production area, the number of pedestrians in video surveillance need to be counted. This research takes video surveillance image as input and uses intelligent video surveillance techniques, realizes the automatic detection, tracking and counting of pedestrian objects in video surveillance. The main contents are as follows:(1) Pedestrian objects detection: this paper uses background subtraction algorithm of Vi Be background modeling to detect moving objects, extracts the HOG features of moving objects, so as to design the far, near and middle distance SVM linear classifiers to detect the pedestrian objects. In order to reduce the impact of light on moving objects detection, as Vi Be background modeling is impressionable of light change cancellation, this paper adopts standard color space(r,g,I) instead of RGB color space, which has certain inhibition effects on light change cancellation. Once an object is similar with background, the thresholds of background subtraction model of the detected pixels of foreground image holes are adjusted, then the foreground detection quality is optimized. When the edge of foreground object image is not continues, this paper combines Canny operator, density of foreground point of margin neighborhood and 7*7 median filter method to repair the foreground image edge. When the background model built in the first frame if contain the target object,the ghost problem will contained in the follow-up test results, this paper introduced a method combining Vi Be background modeling and improved inter-frame difference method to quickly eliminate ghost. Experiment results showed that, to some extent, the improved algorithm increased the detection accuracy.(2) Pedestrian objects tracking and counting: because in Kalman filtering based secondary feature matching for pedestrians tracking and counting method algorithm, the pedestrian objects fusion, fission and static pedestrians scenarios are not considered, this paper introduces an improved Kalman filtering based secondary feature matching for pedestrians tracking and counting algorithm. This algorithm takes the overlapping area of two adjacent frames as main feature, color histogram, shape features of objects as auxiliary features, handles the scenarios when new object appears, as well as objects fusion, fission and static pedestrians scenarios. Experiment results showed that, this algorithm is competent to accurately track and count multiple pedestrian objects.Algorithms introduced in this paper were simulated in opencv2.4.10+VS2010. The images library is from INRIA pedestrians database. Experiment results showed that, these algorithms can realize strong real-time and high accuracy pedestrians counting.