Research And Application Of Automatic Extraction Of Feature Video Segment Based On DSP

Surveillance technology and products replace the staff to continuously monitor the environment such as warehouse,corridor,hall and so on,and storage the monitoring information used to process and analysis.However at present most video surveillance are lake of initiative,they only passively save video data so that more redundant information be saved,that brings great difficulty to video inquiry method at present that mainly based on the artificial to look at the monitor video,at the same time,more and more video streaming data and limited storage space exsist very huge contradiction.Therefore, this paper proposed a method of combining target detection and encoding and decoding technology,Base on target detection drive encoding to extract effective data,reduce the data volume of video stream and redundant information, which is conducive to storage and query.First, On the basis of the research of the common target detection algorithm,as the existing temporal and spatial background difference algorithm needs to calculate the inter and intra characteristic value of pixel of RGB three channels of every frame,and that will result in the bad real-time and high memory usage,so the thsis improved the original algorithm, Algorithm makes full use of gray image and feature information to build robust background model,and then use the background complex degree factor and “first in first out”strategy to update the background and the target detection threshold.experiments in different environments show that the improved method has low memory usage,high accuracy and good robustness.then,the paper proposed binary image connected domain statistical algorithm, the algorithm combines the advantages of the region growing and the linear analysis table that can effectively avoid the repeated scanning and complex data structures, the experimental show that the algorithm is real-time high, can effectively statistics the number of the pixel point of maximum connected field.Then,the paper use video encoding and decoding as the extraction tools,studied H.264 principle, structural frame and realization method.Finally, based on the TI Da Vinci system processor to build the hardware and software platform,On the basis of analysis of its dual core system communication mechanism, the use of SDK suite and H.264 coding principle transplant the object detection and connected domain detection algorithm,and automatic extraction of feature video segmention in a variety of environment,experimental results show that the proposed method can effectively extract the feature video segment, which has a certain application value.