A Detection of Vehicle Velocity Using Deep Learning

In order to keep track of traffic flow, recognize unlawful vehicle classes, flag accidents, and calculate vehicle velocities, Intelligent Transportation Systems (ITS) need to integrate vehicle detection. Even if there is a growing interest in study, this perplexing challenge has not yet been answered and must be addressed. Although hardware options like radars and LIDAR have been suggested, they are excessively costly to run and give operators of traffic monitoring systems with very little data that can be used to take action. The Histogram of Gradients (HOG) and the Gaussian Mixed Model (GMM) are two instances of classical approaches that are employed in software-based solutions. Despite the fact that these methods are employed, they are operationally sluggish and are not suited for real-time traffic detection. As a result, this project will explore and assess a wide variety of approaches to the detection of motor vehicles. In addition, the Spiking Neural Network (SpikeNet) is used to recognize autos from the outputs of highway camera systems in order to carry out video processing procedures and get the required data. In this paper, the anchor frame optimization was done using the YOLOv5s design and the k-means method. This was done under a variety of lighting circumstances. The suggested model was able to attain a mean accuracy performance (mAP) of 97.8 for the daylight dataset as well as 95.1 for the nighttime dataset, in accordance with the results of the simulations and evaluations of the algorithm.