Abstract
The growth in the number of vehicles in the world makes it hard to safely share the environment with pedestrians. Pedestrian's safety is an important task that needs to be granted in the traffic environment. New cars are equipped with advanced driver assistance systems (ADAS) with a variety of applications. Pedestrian detection application is one of the most important applications for an ADAS that needs to be enhanced. In this paper, we propose a pedestrian detection system to be implemented in an ADAS. The proposed system is based on convolutional neural networks thanks to its performance when solving computer vision applications. On the other side, the proposed system ensures real-time processing and high detection performance. The proposed system will be designed by tacking the advantage of building lightweight convolution blocks and model compression techniques to ensure an embedded implementation. Those blocks will guarantee high precision and fast processing speed. To train and evaluate the proposed system, we used the Caltech dataset. The evaluation of the proposed system resulted in 87% of mean average precision and an inference speed of 35 frames per second.