Vehicle Detection and Speed Estimation using Deep Convolutional Neural Network

This research study intends to detect the velocity of moving objects by means of video recording, advanced digital image processing, and deep convolutional neural network. Approach towards the development of an accurate estimation of high-precision speed for vehicles in traffic flows will be proposed. The main contribution of this work is the utilization of the Gaussian Mixture Model (GMM) for background subtraction, which will distinguish the moving objects from the stationary ones. Density- Based spatial clustering of application with noise is applied to cluster and detect vehicles in crowded situations in which they can merge or become separated. Vehicle position prediction and error correction made use of the Kalman filter, following vehicle tracks across video frames. High accuracy was captured estimating the speed without occlusions or sudden changes, and Optical Flow was also used for motion vector fields, where it would be able to capture dynamics in vehicle motion, for example, when changing lanes or turns. The architecture is modular; it consists of preprocessing, detection, track processing modules, and speed computation modules with well-defined communication protocols in between them. This makes it relatively easy to upgrade later with additions like vehicle classification or anomaly detection while keeping a strong data flow and intermediate result refinement intact.