An objective method for pedestrian occlusion level classification

Pedestrian detection is among the most safety-critical features of driver assistance systems for autonomous vehicles. One complex challenges that partial occlusion, where a target object only partially available to sensor due obstruction by another foreground object. A number current pedestrian benchmarks provide annotation occlusion assess algorithm performance in these scenarios, however each benchmark varies greatly their definition occurrence and severity occlusion. In addition, level methods contain high degree subjectivity human annotator. This can lead inaccurate or inconsistent reporting an algorithm’s occluded pedestrians, depending on which used. research presents novel, objective method classification ground truth annotation. Occlusion achieved through identification visible keypoints use effective 2D body surface area estimation. Experimental results demonstrate proposed reflects pixel-wise pedestrians images all forms including challenging edge cases such as self-occlusion, truncation inter-occluding pedestrians.