Human-Targets Tracking Based On Histogram Of Oriented Gradient And Support Vector Machine

The one of key problems for autonomous mobile robot is human targets detection and tracking, it is widely used in visual surveillance systems, service industries and human-computer interaction aspects. Traditional object tracking algorithm based on monocular vision takes the known target color model as a basis, having a high require for the target and background. In addition, the existed tracking algorithms are restricted by many factors in the practical application, such as targets randomly appearing or disappearing,movement multiplicity,occlusion and rotation of targets, and it can not ensure the tracking results when in the relatively complex outdoor environment.The goal is detecting and tracking humans in laboratory and outdoor environment. Our focus is on developing robust feature algorithms that encode image regions as high-dimensional feature vectors that support high accuracy person/non-person decisions. This approach uses low-level appearance and silhouette vectors to detect person, getting rid of traditional color model and keeping invariance for rotation,scale zoom,brightness changing, to test feature descriptor validity we adopt Support Vector Machine as classifier.The core of human target detection is the foreground region observation of objects. Locally normalized Histogram of Oriented Gradients is used as feature descriptor. The HOG descriptor are computed from image gradients, but computed on a dense grid of uniformly spaced cells and use overlapping descriptors standardization for improved performance. Our algorithm of human-targets detection is based on scanning detection window over the whole image at multiple scales and positions, in each position runs classifier for decision, the main function of classifier is coefficients which are trained based on support vector machine.This paper implements multi-human targets tracking in Bayesian framework. Based on exact HOG detection, variable velocity model was established for the target and Kalman Filter was used for durative tracking. Experiment results and analyses on the mobile robot platform show the algorithm's validity and practicability.