P E D E S T R I A N D E T E C T I O N A N D T R A C K I N G
Result sequences from a paper "Robust Pedestrian Detection and Tracking in Crowded Scenes", submitted to Image and Vision Computing.
The first sequence, which we will refer to as the Overhead sequence was set in an indoor scenario with the camera positioned at around 3 metres above the ground. The camera was then orientated back towards the groundplane. The camera rig in this point of view has a limited field of view and due to its proximity with the groundplane it does not encounter significant occlusion problems. The lighting conditions in the scene are stable, but brightly illuminated with a highly reflective ground surface. The sequence consisted of 418 images, captured at a frame rate of approximately 6.5Hz (approximately 1.1 minutes).
The second sequence, which we will refer to as the Corridor sequence was set in an indoor setting with the camera positioned just above 2 metres from the ground. The camera is orientated at 30 degrees towards the groundplane. Again, the lighting conditions are stable, however the scene's illumination is more challenging than that of the Overhead sequence as it is brightly illuminated on one side, and dark on the other side, due to skylights in the corridor. This can cause a lack of texture in those areas, as will be described later. In addition, the scene contains a staircase on the right hand side, where people can descend and ascend at will. The sequence consists of 697 images, captured at a frame rate of approximately 5.3Hz (approximately 2.3 minutes). For these two sequences volunteers were recruited and asked to walk around in front of the camera. No restrictions or instructions were provided as to where people could go, what they could do or what they could wear.
The datasets used in this work can be downloaded here.