Pedestrian detection tracking method based on compressed characteristic

Abstract

The invention provides a pedestrian detection tracking method based on compressed characteristics; the method can import a Gentle Adaboost algorithm with a misclassification cost sensitive mechanism to train positive / negative samples in a training sample set, and comprises the following steps: using a m*n dimension random spacing sparse Toeplitz matrix [phi] Toep as a projection matrix to carrying compressed tracking and dimension reduction for positive / negative samples of a tracking object; commonly inputting compressed and dimension reduced positive / negative sample characteristics and positive / negative sample position weights into a Bayes classifier for training, thus forming a determination function used for distinguishing pedestrian objects from the background; selecting the sample with the maximum determination function value as the new tracking object in the next frame input image, thus detecting and tracking the pedestrian object. The method can improve the pedestrian detection accuracy in a public zone with complex interference factors, thus enhancing the algorithmic robustness; the method is very suitable for tracking and detecting the pedestrian with the head blocked or other abnormal interference conditions.

Description

Technical field [0001] The invention belongs to the field of computer image processing, and particularly relates to a pedestrian detection and tracking method based on compressed features. Background technique [0002] Accurate tracking of target position is a core issue in the field of computer vision. It is widely used in human-computer interaction, video surveillance, and augmented reality. In more advanced tasks such as scene understanding and action recognition, target tracking also plays an important role. [0003] In the prior art, the detection of pedestrians in public areas usually requires training on the positive samples in the positive sample set and the negative samples in the negative sample set to obtain a pedestrian classifier, and the pedestrian area in the real-time collected image is determined by the pedestrian classifier. Judge whether it is a pedestrian area and count the number of people, so as to realize the detection of pedestrians. [0004] At present, mo...

AI Technical Summary

Problems solved by technology

At the same time, when the pedestrian detection method in the prior art deliberately sets the head or shoulders in an abnormal shape, it will miss detection when extracting the pedestrian model and performing detection and tracking, so there are also certain defects.

This kind of defect has a greater potential safety hazard in the occasions where the number of pedestrians is counted or the security inspection is relatively high.

Images