الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 32 |  |  |
| | | from | 32 | | |
Abstract
Object tracking has recently attracted considerable attention from computer vision researchers. The output of object tracking is the basis of many real-world applications in several domains. Object tracking is the process of segmenting an object of interest from a video scene and keeping track of its motion and orientation in order to extract useful information. Although great progress in research has been recently achieved, object tracking is still a very challenging problem. Objects are often being tracked in difficult environments. As a result, object tracking still suffers from a lack of robustness due to many issues including shadows, changing lighting conditions, occlusions, changing appearance patterns of both the object and the scene, non-rigid object structures, and camera motion.
One of the most fundamental challenges of object tracking is the tracking of multiple moving objects. That is, the tracking process is performed on several objects rather than just a single object. It involves keeping track of the positions of a number of independent moving objects and being able to label the tracked objects consistently in different frames of the video. The challenge lies in how to correctly associate multiple objects over consecutive video frames without any confusion. The main goal of multiple object tracking is to find the trajectory of the target objects through a number of frames from an image sequence. It should have the ability to deal with the tracking difficulties incurred by the existence of multiple objects. Although different kinds of approaches have been proposed to tackle this problem, it still has many issues unsolved.
In this thesis, the problem of tracking multi-moving objects is addressed under two different ambient illumination conditions: daytime and nighttime in outdoor environments. More specifically, the problem is investigated through considering vehicle tracking in traffic scenes where the target objects are vehicles. In the daytime, cast-shadow can bring serious problems while extracting moving objects due to the misclassification of shadow points as foreground. In some cases when the shadows stretch, two or more independent objects can appear to be connected together. In addition, in the nighttime, under bad-illuminated condition, the obvious features of vehicles, which are effective for detecting in daytime, become invalid in nighttime road environment. The image has very low contrast and a weak light sensitivity. Furthermore, there are strong reflections on the roads surface, which complicate the problem. The moving reflections of the vehicles headlights can introduce many foreground or background ambiguities.
The main contributions of this research are: (1) a method for removing cast shadow in daytime from vehicles is proposed. The method works by applying a Gamma decoding followed by a thresholding operation and employing the estimated background model of the video sequence. (2) a method for detecting moving vehicles in nighttime is proposed. It identifies vehicles by detecting and locating vehicle lights using automatic thresholding and connected components extraction based on analyzing YCbCr color space. Detected lamps are then paired using rule-based component analysis approach. (3) a comprehensive vehicle tracking system is developed to deal with daytime and nighttime vehicles tracking employing a proposed day/night detector to the scene to determine the suitable technique. (4) a video dataset containing both urban and highway scenes in nighttime is self-collected and prepared to be useful for further research.
Finally, extensive experiments were carried out on different benchmark datasets in order to evaluate the proposed methods. The results demonstrate that the proposed methods outperform the state-of-the-art techniques. The shadow removal method achieves an average shadow detection rate equal to 93.4% (with improvement rate about 4% with respect to the counterpart methods) and an average shadow discrimination rate equal to 88.2% (with improvement rate about 1%). In addition, the method for detecting the moving vehicles at nighttime achieves average detection rates equal to 96.27% and 95.76% for both highway and urban scenes respectively. Finally, the day/night detector achieves an average recall equal to 96.9% and 94.4% for “daytime” and “nighttime” scenes respectively. In addition, it achieves an average precision equal to 95.6% and 96% for “daytime” and “nighttime” scenes respectively.