Traffic scene perception (TSP) aims to real-time extract accurate on-roadenvironment information, which in- volves three phases: detection of objects ofinterest, recognition of detected objects, and tracking of objects in motion.Since recognition and tracking often rely on the results from detection, theability to detect objects of interest effectively plays a crucial role in TSP.In this paper, we focus on three important classes of objects: traffic signs,cars, and cyclists. We propose to detect all the three important objects in asingle learning based detection framework. The proposed framework consists of adense feature extractor and detectors of three important classes. Once thedense features have been extracted, these features are shared with alldetectors. The advantage of using one common framework is that the detectionspeed is much faster, since all dense features need only to be evaluated oncein the testing phase. In contrast, most previous works have designed specificdetectors using different features for each of these objects. To enhance thefeature robustness to noises and image deformations, we introduce spatiallypooled features as a part of aggregated channel features. In order to furtherimprove the generalization performance, we propose an object subcategorizationmethod as a means of capturing intra-class variation of objects. Weexperimentally demonstrate the effectiveness and efficiency of the proposedframework in three detection applications: traffic sign detection, cardetection, and cyclist detection. The proposed framework achieves thecompetitive performance with state-of- the-art approaches on several benchmarkdatasets.
展开▼
