Feasibility Study on Development of a CMOS Vision Processor for Vehicle Tracking

Abstract

Vehicle tracking is an important area of intelligent transportation systems (ITS) technology, which could be applied in a wide range of transportation applications. Tracking typically needs to monitor real-time vehicle movements, and thus real time tracking is highly desirable. However it is well known that vehicle tracking processes are computationally very intensive. In the past, regardless of different algorithms employed in vehicle tracking, they have been implemented using software approaches, e.g., FPGA (Field Programmable Gate Array), microcontroller or embedded micro-processor, and PCs. While software approaches have an advantage of flexibility in implementation and future modifications, its long computational time often prevents real-time vehicle tracking from high resolution spatial or temporal data. It is well known in the area of VLSI (Very Large Scale Integrated) circuit design that a customized and dedicated hardware implementation of any algorithm minimizes its computational time. This gives us the motivation for direct implementation of tracking algorithms in hardware (i.e., device level), whether it is a partial or full implementation, to enhance real-time operation. The goal of this seed project is to investigate the feasibility and related issues in developing a tracking system with a new tracking algorithm based on vehicle motion detection, which is implemented in hardware whenever possible so that the computational time for tracking is minimized. The proposed overall tracking system consists of two parts. One part is the hardware, more specifically, a CMOS (Complementary Metal Oxide Semiconductor) hardware processor which is mainly responsible for vehicle motion detection. The other part is the software, for example an FPGA or micro-controller which is responsible for analyzing the data transmitted from the hardware and properly associating vehicles for tracking. The main computational time saving for the tracking process comes from the hardware part since the core of the new tracking algorithm, motion detection, is run on a dedicated hardware for that particular purpose. The proposed tracking algorithm is simulated in MATLAB and tested on traffic images captured from an intersection. It is found that vehicle movements can be accurately identified in spite of some noisy motion. Also, in this project, we estimate the computational time for the tracking algorithm in hardware implementation and discuss high-level hardware designs for actual implementation of the tracking algorithm.