The paper "Human Detection Using Oriented Histograms of Flow and Appearance" by Navneet Dalal, Bill Triggs, and Cordelia Schmid presents a method for detecting humans in videos, particularly in challenging scenarios with moving cameras and backgrounds. The authors develop a detector that combines motion-based descriptors, specifically oriented histograms of differential optical flow, with appearance descriptors based on Histogram of Oriented Gradients (HOG). These motion features are designed to capture relative limb movements while being robust to camera and background motion. The combined detector is tested on various databases, including a challenging test set from feature films, and shows significant improvements over static appearance-based detectors, reducing false alarm rates by a factor of 10. The paper also discusses the evaluation of different motion coding schemes and the impact of parameter settings on performance. The detectors are trained using a combination of static and dynamic image sets, and the results demonstrate the effectiveness of the proposed method in detecting humans in complex video scenes.The paper "Human Detection Using Oriented Histograms of Flow and Appearance" by Navneet Dalal, Bill Triggs, and Cordelia Schmid presents a method for detecting humans in videos, particularly in challenging scenarios with moving cameras and backgrounds. The authors develop a detector that combines motion-based descriptors, specifically oriented histograms of differential optical flow, with appearance descriptors based on Histogram of Oriented Gradients (HOG). These motion features are designed to capture relative limb movements while being robust to camera and background motion. The combined detector is tested on various databases, including a challenging test set from feature films, and shows significant improvements over static appearance-based detectors, reducing false alarm rates by a factor of 10. The paper also discusses the evaluation of different motion coding schemes and the impact of parameter settings on performance. The detectors are trained using a combination of static and dynamic image sets, and the results demonstrate the effectiveness of the proposed method in detecting humans in complex video scenes.