A method is proposed for detecting pedestrian candidate regions by applying a saliency algorithm model to images taken with in-vehicle monocular camera, for use as element technology in a pre-crash safety system. Pedestrians have diverse modes: appearance depending on their clothes, hairstyles and objects carried, etc., which makes it difficult to define pedestrians.A new method is developed that narrows down the scope for identification of pedestrian candidate regions by utilizing a saliency algorithm model: that is able to define the road surface and thereby identify non-road surface regions. It was also found that pedestrians in the roadway could be consistently identified as attention regions by applying several measures to counter the complex lighting changes and difference in vision that occur as the camera travels. The effects of these methods were verified in tests using real driving images. For the covering abstract see ITRD E134653.
Abstract