Abstract
The performance of a generic pedestrian detector varies based on the data fed to it; when applied to a specific scene, its performance degrades dramatically, which require the detector to be fed with the specific target in mind so that it can produce the desired predictions and detect for the user the specified target. In this paper, I propose to feed the automated specialization of a scene-specific pedestrian detector, with multiple sources from pictures to even videos beginning with a generic video surveillance detector, however manually marking samples to ease the process, as the knowledge accumulated from the master program is still insufficient to produce high-end automated sample marking for the detector. The key idea is to consider a deep detector as a feature that produces a perception of the likelihood of a pedestrian being detected in the target. The system then will be fed with the manually marked samples to enhance its performance and the usage of an already existing system using the Monte Carlo sequential filter system. There has been the implementation of the pedestrian detectors in China, where it showcased the different patterns, the detector can classify and assess whether a pedestrian is present within the testing data or not. The project is truly fascinating as it shows how a machine can learn when fed with the right data and produce sensible results that lead to human renovation and up their living standards by decreasing the number of accidents related to pedestrians affecting the overall rate of accidents. “Many real-world data analysis tasks involve estimating unknown quantities from some given observations” as addressed by the authors within their report on Monte Carlo methods (Doucet A., de Freitas N., Gordon N.). In order to compute rational approximations, it is also important to follow numerical techniques. The techniques of Monte Carlo method (MCM) are powerful tools that allow us to achieve this objective (Andrieu C., Doucet A., Punskaya E.).
Publication Date
4-2021
Document Type
Master's Project
Student Type
Graduate
Degree Name
Professional Studies (MS)
Department, Program, or Center
Graduate Programs & Research (Dubai)
Advisor
Sanjay Modak
Advisor/Committee Member
Ehsan Warriach
Recommended Citation
Alolama, Abduljalil, "Deep Learning of Scene-Specific Classifier for Pedestrian Detection in Dubai" (2021). Thesis. Rochester Institute of Technology. Accessed from
https://repository.rit.edu/theses/10809
Campus
RIT Dubai
