Research And Realization Visual Tracking Algorithm Of Fast-Moving Vehicle Based On Android Platform

Visual tracking of the fast-moving vehicle is an advanced technology which has broad application prospects, and the aim of study is to identify fast-moving vehicles and track time in a video sequence or live video. Traditional visual detection algorithm for fast-moving vehicles, owing to the uncertain speed or vehicle attitude change, often leading to failure of the vehicle tracking. Therefore, aiming at the vehicle tracking problem with different speeds, vehicle stance, light intensity, and partially occluded scenarios, the research of fast-moving visual tracking based on the Android platform is carried out, and the main research contents are as follows:(1) Research on principle of image filter based on optical flow technology. In the process of vehicle in a fast-moving, velocity vector at every point on the body should be the same. The noise points exit in this scene which is of inconsistent with the same velocity vector will be filtered based on this principle. It provides high quality images for detection and tracking, and reduced light flow tracing track drift. Comparing with the effects of traditional linear and nonlinear filters,the effect of optical flow filter is better.(2) Research on fast-moving vehicles detect technology which based on cascade classifier. Applying adaboost framework on three easy acquisitions of weak classifiers for intensive training, a strong classifier is obtained. For each photo,a 13-bit binary code is obtained through the classifier. The probability of this portfolio belongs to the target vehicle is obtained in order to test the classification and identification of samples. This method improve the vehicle detection accuracy at real-time, and provide initial template for tracking module after a full block.(3) Research on vehicle tracking based on optical flow of feedback. The position error between the feature point in the previous photo and the one which is rendered by the feature point in the current photo through backward calculate in the previous photo is got. The key feature points in the current photo are filtered by the difference between the feedback error and the error threshold. They are used to predict the position of the key point in the next frame. By this way the tracking drifts are reduced. Due to the existence of feedback, this method can improve the accuracy of tracking, and reduce key rate drift caused by failure.(4) Research on mobile vehicle tracking algorithm which is based on online learning. The algorithm uses P-N online learning methods, combine a fast vehicle detection and tracking and synchronization, and it assess the accuracy of tracking and detection. It judges whether tracing is on to the next frame or re-training set to update the classification and tracking device.The experimental results based on Android platform show that the method which P-N expert use to combine detection algorithm based on cascade classifier with tracking algorithm based on feedback optical flow implement the complementary of tracking and detection. This algorithm shows good tracking robustness and high accuracy, and it is a good solution to vehicle tracking with the change of speed, attitude, illumination and occlusion.