Human Detection using a Cascade of HOGs   


We integrate the cascade-of-rejectors approach with the Histograms of Oriented Gradients (HoG) features to achieve a fast and accurate human detection system. The features used in our system are HoGs of variable-size blocks that capture salient features of humans automatically. Using AdaBoost for feature selection, we identify the appropriate set of blocks, from a large set of possible blocks. In our system, we use the integral image representation and a rejection cascade which significantly speed up the computation. For a 320 × 280 image, the system can process 5 to 30 frames per second depending on the density in which we scan the image, while maintaining an accuracy level similar to existing methods.



For each level of the cascade we construct a strong classifier consisting of several weak classifiers (linear SVMs in our case). In each level of the cascade we keep adding weak classifiers until the predefined quality requirements are met. In our case we require the minimum detection rate to be 0.9975 and the maximum false positive to be 0.7 in each stage. The training process took a few days using a PC with 1.8GHz CPU and 2GB memory.




Time required to evaluate a 240 × 320 image. Sparse scan refers to the way used in Dalal & Triggs experiment. Dense scan is using a smaller step-size to produce more hypothesis in a testing image. Our method is much faster than the Dalal & Triggs algorithm, and the gap increases as the scan density increases. (For the Dalal & Triggs case, the reported results are based on our implementation of their algorithm). The Rectangular features are the fastest to compute but its accuracy is not as good.




Results using a dense scan and no post-processing was applied to the images


Qiang Zhu, Shai Avidan, Mei-chen Ye, Kwang-Ting Cheng. Fast human detection using a cascade of Histograms of Oriented Gradients. IEEE Computer Vision and Pattern Recognition 2006(CVPR 2006), June 17-22, NYC, USA.

There is a link to the database we used in this project. Thanks Navneet Dalal and Bill Triggs for releasing their database.
<back to research>