Intelligent Algorithms For Video Based Multi Type And Multiple Vehicle Detection And Tracking

Abstract

Multiple Vehicle Detection is a promising and challenging role in Intelligent newlineTransportation Systems and computer vision applications. Most of the existing methods detect multiple vehicles with bounding box representation and fails to trace the location of vehicles. However, the location information is vigorous for several real-time applications such as the motion estimation and trajectory of vehicles moving on the road. In this thesis, the proposed methods namely Improved VIsual background Extractor is used to extract the background and foreground information by first obtain the clusters of foreground and background respectively using mean shift clustering on the background and foreground information; Second, initialize the S/T Network with corresponding image pixels as nodes (except Source S/T Sink node); Calculate the data and smoothness term of graph; Finally, use max flow/minimum cut to segmentation S/T network to extract the motion vehicles on road. An Advanced deep learning method namely Enhanced You Only Look Once version3 which is used to detect the multitype and multiple vehicles by objectiveness score for each bounding box using logistic regression and calculation of using cost function. More precisely, tracking is to find the trace of the upcoming vehicles using a newlineCombined Kalman Filtering Algorithm and Particle Filter techniques by the segments newlinewhich are allocated to the hypotheses are implemented to determine the measurement vector which is in turn used to update the Kalman Filter (segments with association). To improve the tracking results, further, the proposed technique namely Multiple Vehicle Tracking Algorithms by finding trajectory value of using Mode-matching filtering with Extended Kalman Filters and tested with the different weather conditions such as sunny, rainy and fog in input videos of 20 frames per second. The experimental results are tested with the ten different input videos from private datasets and two benchmark datasets KITTI and DETRAC. The six high- level features. newline