Research On Pedestrian State Detection And Forecasting Method In Video

Extensive pedestrian state forecasting in video scenario combines imageprocessing, pattern recognition, statistics, psychology et al.,which is one of thefocuses of current study. The forecasting techniques utilize image processingtechniques of computer science to get the current status of pedestrian. Proper mathmodel is used to gather state statics and forecast future scenario state.In this paper, pedestrian state analyzing method is studied in many aspect. Firstly,flow field is introduced and a pedestrian state detection method based on that isproposed. Especially, data gathering and pre-processing techniques are elaborated.Pedestrian state forecasting model in video is the emphasis of this paper. CellularAutomata(CA) is chosen as the foundation of forecasting. An improved Inner-gridParking Cellular Automata model (IPCA) is proposed due to the shortage of CA indescribing pedestrian motion information. After that a standard micelle-collisionbased state detection method is proposed according to the feature of extensivepedestrian. The method use k-Means algorithm segmenting feature points into severalmicelle sets. State of current scenario is predicted by analyzing relationship betweeneach micelle. The model deduce state of scenario from current video frame byanalyzing historical forecasting states generated it. A feed-back method is introducedto reflect changes of scenario in time by revising prediction output with currentdetected information so that the accuracy rate is improved. Public dataset is used astest data in state forecasting. The results of the model is compared with otherdetection methods to demonstrate excellent performance.Finally, the model is implemented by utilizing OpenCV and Qt libraries in C++.The correctness and effectiveness of IPCA model is verified by simulation process.The result of the prediction is saved as the foundation for the intention of featurevideo analyzing use.