Multiple objects detection on street using Hmax features and color clue

This paper presents a model for multiple objects detection using Hmax features and color clue for combined detection of interest objects on street. On this model, input image is converted to grayscale and scaled to the appropriate size. On Greyscaled image, hmax features will be extracted for each interest objects as in [1] with tuning parameter to fit with each object. These Hmax features will be passed to correlative Hmax detector of each object. To reduce time for detect a large image, instead using a model with full image size like [1], image is scaled and detected at fit size and at potential positions. Detected positions will be found for 7 objects on the same image for color and none-color image. If image is color, color clue will be extracted to filter results and adjudge segmentation of some texture objects. Color clue will be extracted by filter threshold of some texture objects, these clues will be used to fine-tune detected position of texture objects. These color robust for [1] ability segment for texture object by color without using segmentation tool in [7]. Model is tested for interest objects on the same image with different image sizes. In [7], time for detecting for a image with size 960x1280 is approximate 300s for extracting full set hmax features on an object with CPU 2G. Time for model of paper run on GPU is approximate 20s for detecting an object on full image. This paper also try to implemented thing that [7] and [1] has not implemented yet: combining single detectors in one detection system. This combining will help detection systems reduce complex and improve speed easier. For detecting presence or absence of objects on street, the result of our model is 89.79% on Streetscence database [7], and 87.5% on random database.