A Vehicle Detection Scheme for Heterogeneous and Lane-Less Traffic

Abstract

A traffic management system (TMS) is essential to manage traffic congestion, save fuel, save travel time, and enhance user safety. A vehicle detector is a vital component for building such TMS. Different types of vehicle detectors are developed for homogeneous and disciplined traffic, but they are not suitable for heterogeneous and lane-less traffic conditions, which is common in developing countries. The proposed work presents a dimension based vehicle classification technique. At first performance of vehicle classification based on dimensions is verified analytically. Thereafter, a vehicle detection scheme for heterogeneous and lane-less traffic by extracting a binary image from the sensor nodes is proposed. The binary image is recorded based on the occupancy status at respective sensor zone as logic 1 or 0. Vehicle classification and its corresponding speed is estimated from the obtained binary images. Initially, the proposed work is validated by a photodetector based set-up. Then, with the help of scaled-down 2D vehicle models, a binary image is obtained. Photo-detector based set-up serves dual purpose, i.e., it can be used for vehicle to infrastructure (V2I) communication using visible light and vehicle detection. Moreover, in the real scenario, proposed method is validated by generating a binary image from virtual loops in a video recording. To enhance its effectiveness a convolutional neural network (CNN) based foreground detection method is applied. But, camera based systems efficiency reduces by environment conditions. To overcome this issue, an array of micro-LiDARs is designed and implemented. The proposed method requires less storage, less bandwidth, less computation complexity, and easily can be implemented. The detection accuracy of 98% is observed while extracting data from video and 91.3% while using micro-LiDARs. The proposed works are compared with existing techniques. Finally, the obtained data is recorded remotely and displayed it on customized web map. A traffic management system (TMS) is essential to manage traffic congestion, save fuel, save travel time, and enhance user safety. A vehicle detector is a vital component for building such TMS. Different types of vehicle detectors are developed for homogeneous and disciplined traffic, but they are not suitable for heterogeneous and lane-less traffic conditions, which is common in developing countries. The proposed work presents a dimension based vehicle classification technique. At first performance of vehicle classification based on dimensions is verified analytically. Thereafter, a vehicle detection scheme for heterogeneous and lane-less traffic by extracting a binary image from the sensor nodes is proposed. The binary image is recorded based on the occupancy status at respective sensor zone as logic 1 or 0. Vehicle classification and its corresponding speed is estimated from the obtained binary images. Initially, the proposed work is validated by a photodetector based set-up. Then, with the help of scaled-down 2D vehicle models, a binary image is obtained. Photo-detector based set-up serves dual purpose, i.e., it can be used for vehicle to infrastructure (V2I) communication using visible light and vehicle detection. Moreover, in the real scenario, proposed method is validated by generating a binary image from virtual loops in a video recording. To enhance its effectiveness a convolutional neural network (CNN) based foreground detection method is applied. But, camera based systems efficiency reduces by environment conditions. To overcome this issue, an array of micro-LiDARs is designed and implemented. The proposed method requires less storage, less bandwidth, less computation complexity, and easily can be implemented. The detection accuracy of 98% is observed while extracting data from video and 91.3% while using micro-LiDARs. The proposed works are compared with existing techniques. Finally, the obtained data is recorded remotely and displayed it on customized web map.