Traffic Anomaly Detection and Annotation Using Hybridization of Deep Learning Method and Haar Cascade Classifier

The increase in population in most countries has raised the issue of effective traffic monitoring and control. Nowadays, analyzing urban traffic conditions is greatly dependent upon data-centric analytics. Video is a most crucial form of data that is captured by roadside cameras to acquire traffic data. Analyzing such videos reveals several traffic conditions. In this context, this paper discusses a sequential hybridization of the Convolutional Neural Network Classifier (CNN) and Haar Cascade Classifier (HCC). The CNN model is used for extracting the features of the vehicles, subsequently, HCC classifies the vehicle type. This hybrid model is used for the identification of four types of vehicles: Car, Bus, Minibus, and Truck. Along with this, it indicates whether a particular video frame is anomalous or not based on the percentage of the area of a frame (denoted as r%) being covered by the vehicles. We consider 7 models of hybridization with HCC. The results show that CNN+HCC outperforms the other models in identifying the vehicles.