Content area
Full Text
Abstract - We propose a multi-level fusion scheme for target detection using camera and radar sensors. For the proposed scheme, the radar sensor provides the target track information of the range, velocity, angle, and track ID. This data is applied during the vision processing step as the ROI (region of interest) where the target may exist. Next, for feature-level fusion, the Doppler spectrum of the ROI is provided to the sensor-fusion-based target classifier. In the classifier, we then determine the class of the target using an image database and a Doppler pattern database. In the experimental results, we verify the proposed processing scheme using a 24 GHz FMCW transceiver with a single antenna.
Keywords: sensor fusion, radar sensor, automotive radar
1 Introduction
While the roles of vehicles included only transportation in the past, safety has become an important requirement of vehicles. Thus, various ADASs (advanced driver assistant systems) have developed, and many such applications have been implemented into commercial vehicles recently. For the effective operation of an ADAS, the sensors are very important, with the most popular sensors being camera and radar sensors [1].
Because the camera is a feature-based target detection sensor similar to the human eye, it has been applied in applications such as LDWSs (lane-departure warning systems), LKA (lane-keeping assist systems), AVM (around view monitoring), and TRS (traffic sign recognition). However, camera sensors depend on external conditions such as the level of illumination or the weather conditions, and the detection accuracy of range and velocity are very low.
On the other hand, radar sensors are robust to external conditions, and the detection range and velocity are highly accurate. Thus, radar sensors have been applied in relation to ACC (adaptive cruise control), LCA (lane-change assistance), FCW (forward collision warning), and BSD (blind spot detection) systems. However, the angle position of the target is estimated at a low resolution, and the target classification performance is limited in these systems at present [2].
Thus, the concept of 'sensor fusion', referring to a combination of radar and cameras, has recently attracted the attention of researchers who attempt to enhance the target indication performance capabilities of automotive applications. Especially considering that the Euro NCAP (New Car Assessment Program) requires the detection of pedestrians and bicycles, sensor...