Detection and classification of vehicles from heterogeneous traffic video data collected using a probe vehicle
Dhanalakshmi Renganathan, Leena Mary, Anu George · 2018
On typical roads in India where traffic are heterogeneous, road planning and management become crucial. Vehicle detection and classification is useful for traffic survey, planning, signal time optimization, and surveillance. A vehicle with a video camera placed at the top can be used for collecting video of the traffic for a two lane undivided road. The camera is placed at an angle of approximately 45 degrees. Both audio and video are used for detection and classification. The data collected is later processed for detecting and categorizing vehicles in the traffic into seven categories namely, one vehicle, more than one vehicle, no vehicle, two wheeler, three wheeler, cars and heavy motor vehicle. The presence of vehicle is identified by analyzing the short time energy of the audio peaks and the process of adaptive background subtraction. Selection of detected vehicles is carried out using Speeded up Robust Features (SURF) matching algorithm. Histogram Oriented Gradient (HOG) features of selected vehicles is extracted and classification is performed in two stages, using Support Vector Machine (SVM) classifier.