Research On Visual Tracking Via Hough Forests Based On CUDA

Motion Target Tracking technology have broad application prospects in military and civilian areas. However, the target tracking system always work in clutter environment which includes partial or completely occlusion, appearance change, pose change, illumination change, such factors leads a integrate trouble for tracking technology.This paper used multiple feature fusion, CUDA programming techniques combined with random forests theory. We propose an more accurate and robust tracking algorithm based on Haar Hough forests. The main work and contribution of this paper is as follows:Firstly, we propose an effective and robust tracking algorithm based on Haar Hough forests. The algorithm adopts the LBP feature because LBP is a feature that can be used to characterize the texture feature information. Besides, we proposed a new strategy of splitting function during the training phase, In the training phase of the original Hough forests, decision function randomly selected two points where the pixel values, representative of this information is not enough, we consider the use of the Harr-like response instead, this can get more neighborhood information near the selected points.Secondly, Implementation of Hough forests tracking algorithm based on CUDA. We use CUDA parallel programming techniques to build and test Hough forests, experiments show that CUDA technology effectively improve the speed of training of the forests, and achieved a high speedup.Apply this paper’s algorithm on single object tracking sequences, and compared the tracking accuracy with other five state of the art single object tracking algorithms(Frag, MIL, IVT, PN, VTD), experiments show that our approach is superior to the other five algorithms on average center location error. We have achieved accurate and robust tracking.