Vehicle detection for thermal vision-based traffic monitoring system using principal component analysis

Machine vision is a popular technology used in Traffic Monitoring System (TMS) to detect vehicles in the traffic scene. Recently, thermal vision provides an alternative machine vision for the TMS since it demonstrates good vehicle detection accuracy, especially under night condition. Under the thermal vision, the vehicles appear almost similar in the daytime and nighttime, even though the illuminations of the traffic scene are significantly different. This vision effect motivates the development of a single vehicle detection algorithm that works in both of the illumination conditions. It is a challenge that has existed for decades. In this paper, a framework for thermal-vision-based TMS is proposed. Histogram of Oriented Gradients (HoG) is a feature descriptor used to recognize the vehicles on the road. The similar appearance of the vehicles under the thermal vision allows the use of single adaptive reference descriptor in vehicle detection for each lane of the road, in which the reference descriptor is generated and optimized with Principal Component Analysis (PCA). In both the daytime and nighttime, the proposed TMS framework has demonstrated high vehicle detection accuracies under the thermal vision