Design And Implementation Of A DSP-based Embedded Intelligent Traffic Surveillance System Front-end Equipment

Increasing congestion on urban streets and freeways boosts the demands of city traffic law enforcement and traffic management. To ensure the accuracy of evidences of traffic violence events, and implementation of uninterrupted traffic surveillance, management and control, it is necessary to bring in automatic traffic manage and surveillance system. The rapid development of computer hardware, digital image and video processing technology make the initiation possible. Currently there have been some urban traffic surveillance applications, but they still have defects such as expensive, poor reliability and isolation of discrete systems. It is required to build an integrated and information sharing intelligent traffic surveillance and management system.The project focuses on urban intersection surveillance system, which belongs to intelligent traffic surveillance systems and covers comparatively most complex scenes in road traffic surveillance. Therefore, it is easy to adapt the intersection surveillance algorithms to the other applications. Due to high costs and low reliability of conventional PC based systems, we proposed a DSP based intelligent traffic surveillance system. This thesis consists of hardware and software. The hardware section includes schematics capturing, printed circuit board layout; the software section covers design of system software architecture, bottom-level operating system porting and introduction of vehicle detection algorithm.In full consideration of application demands, the overall system design is introduced, including layout of system cameras, distribution of video processing software and hardware, processor selection, etc. The detailed design expands to the DDR2 memory, flash memory, peripherals and ports, as well as system power supply. The 594 MHz TMS320DM6446 and utilization of DDR2-533 memory chips can be characterized as high-speed digital circuits, in which signal integrity (SI) issues such as reflection and crosstalk cannot be neglected. A minimum 6-layer PCB is required to fulfill the normal operation of the circuit. We bring Cadence SPB 16.2 and Hyperlynx 8.0 in the design process to aid the system schematic capturing, layout and SI analysis, and high-speed differential signal and DDR2 layout are among the most significant SI issues. Moreover, EMC and EMI problems are fully taken into consideration in the design.On the basis of overall software architecture, the thesis finished porting of user bootloader, embedded Linux operating system and file system, all of which comprises a complete embedded software and hardware development platform. In the software application level, a temporal-mean background modeling and subtraction algorithm is proposed to implement vehicle detection, in which Ostu's threshold method is used to optimize the process of image binarization. Blob analysis in applied after the separation of foreground and background. With the blob algorithm, it is easy to classify road vehicles and remove the unwanted objects like pedestrians. Moreover, Kalman filter is added to remove noise and optimize the tracking stability. The hardware design succeeds at the first attempt, coupled with successful low-level software porting, set up a complete development platform on the basis of TI Davinci technology; the vehicle detection algorithm is programmed in C language, and well prepared for later porting attempts.