Automatic Vehicle Detection and Tracking using Cascade Classifiers and CNN-based Multi-scale Feature Extraction

This article introduces a novel automatic multiple vehicle detection and tracking framework that combines successfully computer vision to partial differential equation (PDE) - based models. Its detection component applies a Haar Cascade Classifier and an Active Contour-based segmentation using a nonlinear PDE-based level-set scheme, to locate the vehicles in the traffic video. The proposed tracking by detection method consists of a deep learning-based multi-scale analysis of the detected vehicles, using a nonlinear second-order reaction-diffusion based scale-space representation, and a vehicle matching process. The simulation results that are finally discussed here illustrate the effectiveness of the proposed approach.