Traffic monitoring system , target tracking and behavior recognition
|School||Central South University|
|Course||Traffic Information Engineering \u0026 Control|
|Keywords||Target tracking Mixed Boosting Algorithm A mixed target detection model Trajectory analysis Behavior recognition|
Intelligent Traffic Monitoring System for automated detection of traffic incident, pedestrian or vehicle intelligent monitoring, and can adapt to the needs of practical application. Thesis goal Intelligent Traffic Monitoring System detection, target tracking, and target behavior analysis to understand the key issues in the three aspects to conduct an in-depth study and propose new solutions, mainly reflected in the following aspects: (1 most of the use of a single model) for the current target to detect the presence of problems such as the high rate of false alarms, light sensitive, dynamic scene robustness and poor, a hybrid motion detection model, the model will be insensitive to changes in illumination target detection motion detection model and dynamic scene changes fast tracking capability fusion, fusion strategy to eliminate missed and false alarms in the detection process. Last fast moving target detection method is proposed to reduce the computational model, coupled with the fusion of the two models have better real-time characteristics of the hybrid model still have some real-time. (2) description of the target tracking process, proposed moving target tracking algorithm based on multiple feature selection. Combination of RankBoos and AdaBoost to build mixed boosting algorithm Select features based on target information and background information to establish the characteristics Sort classifier adaptively updated constantly, and in the process of tracking. Kalman filter is used to rough prediction of the target area, and then use the Sort classifier combination Mean-shift algorithm to complete target tracking. The algorithm can be based on a different target and background information, the adaptive feature selection, it is very advantageous to overcome the problems of present in the scene illumination, interference, occlusion. (3) movement behavior recognition method based trajectory analysis. Trajectory tracking behavior patterns by using clustering methods to learn the sport mode trajectory reference sequence. Then the track as a time sequence, the use of dynamic time Reformed (DTW) no limit on the length of time series characteristics, the DTW and the K-nearest neighbor algorithm used in combination to be identified track with the reference sequence template trajectory matching, the matching process, DTW lower bounds excluded a large number of similar trajectories, in order to accelerate the speed of matching, and then identify the target state of motion. Experimental results show that the target detection and tracking algorithm of target detection and stable tracking, behavior recognition method based on the trajectory analysis of the movement at the crossroads of pedestrian turn left, turn right, forward, U-turn reached higher recognition rate.