Study Of Passenger Flow In Vision Detection System Based On DSP

Developing the passenger flow detecting technique is a new area in vision detection field, which is an effective way to resolve traffic problem. It will help monitoring transportation performance, improving the plan, scheduling and operations of the service. The aim of analysing the video sequence which captured from camera is to detect and track the passengers. The disposal of passenger flow statistic in vision detection is given. The main research work of this dissertation is to take full advantages of DSP to process the images, analyze the key techniques of vision detection, and make the real-time system come true finally.After the three different common target detecting algorithms are analyzed, the background subtraction is presented, which includes whole of the moving information between the two different images. A new algorithm of moving target existing detection based on moving information statistic is presented in this dissertation.Compared with the other four different commom target segmentation alogrithms, an effective thresholding method is presented, which is very suitable for the bus environment. The fixed threshold combined with the adaptive threshold from the image diversification is proposed, which is based on its own grayscale.During the tracking process, an improved tracking algorithm based on target-chain is proposed. The target-chain which includes partition of the tracking area and data structure of every target is established, first. Then, in term of moving target tracking, the character-based tracking approach is adopted and a kind of cost function is improved in tracking and counting algorithm.At last, a moving target detection and tracking system on the ADSP-Blackfin533 is designed and completed in this dissertation. The result shows that the proposed technique used in the system quickly and correctly, which can be used to realize to count passengers exactly.