Scale-adaptive Object Tracking Based On Pixels Feature-weighted

Visual tracking is an important task within the field of computer vision.With the widely used applications in intelligent video surveillance system,and intelligent drive,object tracking technology is facing increasingly more challenges.An effective method of object tracking using weighted pixel features is proposed to deal with multiple complicated tracking situations,such as target movement,rotation,background interference and scaling etc.The algorithm based on Mean-shift tracking.By building the object model,tracking efficiency is significantly improved,scaling adaptive and updating the model is achieved.First,the color feature and location information of the pixels in the target area are used to build the object model.In the target area,each pixel in distinguish between background and the object is very different.It is similar to people identifying fast-moving objects based on significant features.Different pixels play different roles in the model of target recognition.The object model is built in order to make the significant feature of every pixel has a more important role in the target model,and avoid increasing the computational cost from the multiple features fusion.The color features and location feature of the pixels are used as the significant feature.The pixel with significant color features and close to the center position are given greater weight,while the pixels similar to the background in the target region are with lower weight.Then the average weight image is used to estimate the scale variation coefficient.The goal is to adapt to the scale changes of the target.The scale changes constantly with the move of the object,and the change of observation angle or the distance.Predicting the target scale of the next frame is necessary in order to achieve the scale adaptive of the object,and describe the target feature accurately.The ratio of the average weight image is used as the scale variation coefficient.Amend the scale by combining with the initial frame.All previous frames are used to estimate the scale of the next frame.These techniques are applied to guarantee the accuracy of the tracking.Finally,an update model is proposed,which is able to renew the object model and background model.Object tracking is a continuous process,and the targets are often faced with the interference of background.To adapt to the changes in the background,the model must update real-timely.Combining with the temporal context information in consecutive frames,and according to the strength of the background change,the background model updates selectively.To reduce the interference from background in target region,object model is updated also according to the predicted scale of the object,so as to achieve accuracy and stability the tracking process.The experimental results show that the proposed algorithm could make full use of the differences between pixels in the target area,so it can track more quickly and more effectively with strong robustness.