In computer vision, accurate vehicle recognition and tracking is a challenging research topic. Manual surveillance systems are cumbersome, labor-intensive, and inefficient in today's congested traffic environment. Automated traffic monitoring systems are needed in today's world because they provide potential solutions to these issues. However, there are still some challenges in automating these systems which need to be addressed. One of the challenges is the extraction of foreground from a complex environment that only contains vehicles along with spatial and temporal information. In this paper, a semantic segmentation-based foreground extraction model is proposed to overcome these challenges. After segmentation, a blob detection technique is applied to detect each vehicle. To overcome the issue of vehicle-to-vehicle occlusion we extracted Harris corner points and computed the separation line based on the vehicle geometry coordinates. Furthermore, we introduce an ID assignment mechanism based on VGG-16 extracted features. Based on the detections, the number of cars is also counted for each frame. For tracking, we first applied Farneback optical flow algorithm to detect motion in vehicles. Only those vehicles were passed to the tracking algorithm in which motion was detected. To estimate each vehicle's location in succeeding image frames Kalman filter was implemented. A vehicle reidentification and ID recovery module are also proposed to enhance the tracking outcome. Vehicle trajectories based on vehicle location midpoint were also estimated. Extensive experiments are used to validate the proposed system and achieved a precision score of 0.965 and 0.931 over Roundabout and AU-AIR datasets for vehicle detection. However, for vehicle tracking the precision is 0.948 over the Roundabout dataset and 0.839 for the AU-AIR dataset.
Semantic Segmentation Based Real-Time Traffic Monitoring via Res-UNet Classifier and Kalman Filter
SN COMPUT. SCI.
SN Computer Science ; 6 , 1
2024-12-22
Article (Journal)
Electronic Resource
English
Occlusion handling , Kalman filter , Res-Unet , Semantic segmentation , Blob detection , Detection and tracking Information and Computing Sciences , Artificial Intelligence and Image Processing , Computer Science , Computer Science, general , Computer Systems Organization and Communication Networks , Software Engineering/Programming and Operating Systems , Data Structures and Information Theory , Information Systems and Communication Service , Computer Imaging, Vision, Pattern Recognition and Graphics
Localized Extended Kalman Filter for Scalable Real-Time Traffic State Estimation
Online Contents | 2012
|