Abstract:

A new method for bicycle detection is presented. This method is different from the traditional video detection framework of background model with background update and foreground detection. A window classifier is constructed by histograms of oriented gradients and support vector machine classifier to explore that whether a bicycle exists in the window space or not. The classifier is trained by bicycle and nonbicycle samples to get the classify mode. Accordingly, the method can satisfy the need to detect bicycle under fixing and moving camera conditions. In experiment, 700 bicycle samples and 500 nonbicycle samples were trained to get the classify mode of the window classifier. Bicycle videos include different traffic conditions were tested and results validate the proposed method.

Key words: video detection, histograms of oriented gradients, bicycle detection, support vector machine