Video-Based Bicycle Detection in Underground Scenarios

Automatic surveillance systems are animportant emerging application of object detection algorithms in video. Thenature of such systems implies several requirements on the used algorithms.Also, searching for less usual objects (in contrast to frontal human faces, carmasks, etc.) is required, such as detection of bicycles. It appears thatdetection of such objects cannot be solved by just applying a standardstatistical or other general detector, but by constructing a specializeddetector composed of several standard image processing and object-detectiontechniques combined together ad hoc. A detector of bicycles in video data fromstandard low-resolution CCTV surveillance system is presented in thiscontribution.

Bicycle detection approach covered by thispaper aims to cope with highly-noisy low-resolution data, to use simpleimage-processing methods and to work in real time. Although the method itselfdoes not constitute a generally usable object detector, it covers severalinteresting aspects which can be re-used in tasks similar to the given one.Low-level features extracted from the video used for wheel-candidateclassification are described in detail. The system is applied and evaluated onreal data and the results are discussed.