This dissertation discusses three major issues that are related to traffic monitoring. The underlying features used in this work are adopted all based on the entropy measurement. Entropy is a commonly measurement that can be adopted to describe the degree of disorder in thermodynamics. It is worth noting that a detection zone that has more blocks containing active pixels, its corresponding entropy will be relatively high, even though the amount of actives pixels is small. Based on this useful property, entropy-based features tolerate a wide range selection of the threshold value. Since the detection of entropy is immune to the effect of lighting condition, it can accommodate to different weather conditions, such as sunny days, cloudy days and rainy days.
The first topic discussed in this dissertation is about traffic parameter extraction. Based on an entropy measurement, a number of important traffic parameters, such as traffic flow, space mean speed and traffic queue length, can be determined in real time. On the other hand, since the accurate computation of entropy measurement depends on the distribution of active pixels, thus an efficient background updating is always required at a certain time. In order to obtain the accurate traffic parameters, we also propose a better background updating method. Instead of updating the whole background image, we must update the area of interest. As a result, the entropy value of the area of interest must be calculated frame by frame. Under these circumstances, the proposed method can save a lot of computing time and it makes some real-time applications possible.
The second topic discussed in this dissertation is to develop a new method for real-time vehicle detection and tracking. Based on a number of features, we propose a macroscopic method which is able to perform real-time tracking of moving vehicles on highway. In addition to tracking a normal car running on highway, the proposed method can also track a vehicle performing lane change. This approach consists of two phases: a detection phase and a tracking phase. In the detection phase, we use entropy-based features and the salient edges of a vehicle to check for the existence of vehicles. Then, we use a bounding box to track the targeted vehicle, and use its velocity to estimate the possible location at the next image frame. The, we perform the tracking task based on the extracted entropy features to determine the accurate location of the target vehicle. By conducting a great number of experiments, the et number of experiments, the experimental results demonstrated that the proposed system is useful.
This third issue covered in this dissertation is with regard to the automatic detection of traffic accident. We propose a 2D cellular model based on entropy-based features to describe the dynamic behavior of the freeway traffic. Using the proposed model, the complex traffic of multilane can be easily represented by a dynamic discrete system. Based on a number of important information derived from the above mentioned system, traffic incidents can be easily detected. From the experimental results, we demonstrated the efficiency as well as the effectiveness of the proposed system.||en_US|