This paper presents a human detection method using oriented histograms of flow and appearance. The authors propose a detector that combines motion-based descriptors with appearance-based descriptors to improve detection performance in videos with moving cameras and backgrounds. The motion-based descriptors are derived from oriented histograms of differential optical flow, while the appearance-based descriptors are based on Histograms of Oriented Gradients (HOG). The resulting detector is tested on several databases, including a challenging test set from feature films with wide variations in pose, motion, and background. The combined detector significantly reduces the false alarm rate compared to appearance-based detectors, achieving a false alarm rate of 1 per 20,000 windows at an 8% miss rate. The paper evaluates various motion coding schemes and finds that oriented histograms of differential flow provide the best performance. The motion descriptors are combined with HOG appearance descriptors to create a robust detector that is effective in both static and dynamic scenes. The detector is trained on a combination of static and dynamic data, and tested on two challenging test sets containing over 4400 human examples. The results show that the combined detector outperforms both motion-only and appearance-only detectors. The paper also discusses the use of optical flow estimation methods and evaluates different motion descriptors, including those based on spatial differences and Haar wavelet-like operators. The results show that the best combined detector uses motion descriptors based on oriented histograms of differences of unregularized multiscale flow relative to corresponding pixels in adjacent cells (IMHcd) or to local averages of these (IMHmd). The detector is evaluated on several test sets, demonstrating its effectiveness in detecting humans in videos with moving cameras and backgrounds. The paper concludes that the combined detector provides a robust and effective solution for human detection in dynamic scenes.This paper presents a human detection method using oriented histograms of flow and appearance. The authors propose a detector that combines motion-based descriptors with appearance-based descriptors to improve detection performance in videos with moving cameras and backgrounds. The motion-based descriptors are derived from oriented histograms of differential optical flow, while the appearance-based descriptors are based on Histograms of Oriented Gradients (HOG). The resulting detector is tested on several databases, including a challenging test set from feature films with wide variations in pose, motion, and background. The combined detector significantly reduces the false alarm rate compared to appearance-based detectors, achieving a false alarm rate of 1 per 20,000 windows at an 8% miss rate. The paper evaluates various motion coding schemes and finds that oriented histograms of differential flow provide the best performance. The motion descriptors are combined with HOG appearance descriptors to create a robust detector that is effective in both static and dynamic scenes. The detector is trained on a combination of static and dynamic data, and tested on two challenging test sets containing over 4400 human examples. The results show that the combined detector outperforms both motion-only and appearance-only detectors. The paper also discusses the use of optical flow estimation methods and evaluates different motion descriptors, including those based on spatial differences and Haar wavelet-like operators. The results show that the best combined detector uses motion descriptors based on oriented histograms of differences of unregularized multiscale flow relative to corresponding pixels in adjacent cells (IMHcd) or to local averages of these (IMHmd). The detector is evaluated on several test sets, demonstrating its effectiveness in detecting humans in videos with moving cameras and backgrounds. The paper concludes that the combined detector provides a robust and effective solution for human detection in dynamic scenes.