End-to-end quantum-inspired method for vehicle classification based on video stream

Intelligent Transportation Systems (ITS) are the most widely used systems for road traffic management. The vehicle type classification (VTC) is a crucial ITS task due to its capability to gather valuable traffic information. However, designing a performant VTC method is challenging due to the considerable intra-class variation of vehicles. This paper presents a new quantum decision-based method for VTC applied to video streaming. This method allows for earlier decision-making by considering a few stream’s images. Our method is threefold. First, the video stream is acquired and preprocessed following a specific pipeline. Second, we aim to detect and track vehicles. Therefore, we apply a deep learning-based model to detect vehicles, and then a vehicle tracking algorithm is used to track each detected vehicle. Third, we seek to classify the tracked vehicle according to six defined classes. Furthermore, we transform the tracked vehicles according to a pipeline, consisting of the histogram of oriented gradients (HOG), and principal component analysis (PCA) methods. Then, we estimate the vehicles’ probabilities of belonging to each class by training multilayer perceptron (MLP) classifier with the resulting features. To assign a class to a vehicle, we apply a quantum-inspired probability integrator that handles each frame’s information flow. The unique characteristics of the work we propose, compared to the existing ones, are expressed in the decision-making process, since the former requires a sequence of frames of different sizes, compared to the image-based-decision made by the other methods. Our method outperformed the baseline methods with an accuracy up to 96%.