Research On Target Detection And Tracking Algorithms Based On Neural Network

The analysis technique of moving target based on video involves in knowledge of each fields such as computer vision, the image processing and automation control, which provides information about location, detection, investigation and navigation for weapon equipment, intelligence analysis system, satellite, the ship and missile. Besides, the motion target deception and tracking is core technology in the analysis system of moving target based on video.Numerous researches have done plenty of work and paid close attention to the detection and tracking algorithm. Although many achievements have being obtained, the detection and tracking algorithm still can hardly satisfy the request of complicated scene. With rapid development of moving target tracking algorithm in recent ten years, varies of tracking algorithm have been put forward to cope with distinct scenes. Due to various methods representing the tracked target, the tracking algorithm of monocular camera mainly includes point tracking, silhouette tracking, kernel tracking and structure model trackings and these methods include most target tracking algorithms. Due to random variation of the moving target and environmental conditions, the robustness, accuracy and real-time performance of the detection and tracking algorithm would be influenced.In the aspect of motion detection, this paper study a detection and classification model called BP_Adaboost based on BP neural network and Adaboost algorithm via further research on existing motion detection algorithm. The model is applied to detect target through strong classifier which is composed of multiple BP neural network weak classifiers based on Adaboost algorithm and can raise the accuracy of pedestrian detection. The analysis of experimental results shows that the method can accurately classified detect out moving target and greatly improve the robustness of detection algorithm.In the aspect of target extraction, due to the complicated environmental conditions of target, though commonly adopted target feature such as color, texture, grayscale and edge feature has simple structure of expression, the target feature is easy interfered with the environment and has great influence on the accuracy of detection. To overcome this problem, this paper improves algorithm performance via extracting local feature of target. Corner points of the target obtained based on Harris operator is treated as the sample set of input features,which improves the accuracy of target tracking.In the aspect of moving target detection, the paper mainly studies the moving target tracking based on Camshift algorithm. Due to the uncertainty of motion direction, the Camshift algorithm cannot accurately track the target if moving too fast, which may lead to the loss of tracking. According to the problem, this paper takes the trajectory prediction into consideration and studies a new mothed being able to track moving target,besides, improves the traditional Camshift target tracking algorithm and realizes fast and accurate tracking of moving target.