| Summary: | 碩士 === 元智大學 === 電機工程學系 === 98 === This paper presents a detection algorithm, which can detect and recognize vehicles, motorcycles, and pedestrians on the roads. This method still works efficiently and effectively although occlusions occur. At the training step, we obtain the foreground objects’ feature parameters respectively, and thus these feature parameters can be exploited to recognize vehicles, motorcycles, or pedestrians in our system. According to the Gaussian distribution and the entropy of feature parameters, we obtain the weight value of each feature parameter. At the detection step, we use distance measurement function to obtain the likelihood between the sampling data and feature parameters. Then, we use the likelihood value to find the precise location of the exact object among the several candidates of detection objects. In addition, this paper employs Mean-Shift method for object tracking and records object-moving velocity to differentiate motorcycles or pedestrians.
|
|---|
