Abstract:
Abstract
The problem of pedestrian-vehicle crashes is a major cause of deaths and
injuries in road accidents worldwide. The major factor of such crashes is driver.
However, there is no effective solution for this problem. In this project, we propose a
microcontroller-based pedestrian detection prototype system for this problem. Our
approach is based on image processing and machine learning methodologies to detect
and classify images with or without pedestrians. The prototype contains a camera that
can be installed on the car's dashboard, a microcontroller for computational processes,
and a speaker for alarming the driver. The first phase of this project is building a binary
classifier, which can predict if the analyzed images contain a pedestrian or not. The
Histogram of Gradient (HOG) was used as feature descriptor to encode the processed
images. An initial Support Vector Machine (SVM )classifier was built and trained on 250 positive and 250 negative non-redundant images. A 5 fold cross validation was showed a very high accuracy of the created classifier. In the second phase, we tested our trained model on real-time images. Our system achieved a moderate accuracy and a processing speed of 7 frame per second. The accuracy of the system is dependent on the number of training images. In conclusion, we have shown that microcontroller can be used to successfully implement an SVM pedestrian classification system that can be further optimized to a prototype product. To our knowledge, this is the first attempt to implement a pedestrian detection system using a Raspberry-Pi microcontroller
