This paper presents a comprehensive survey and empirical evaluation of moving shadow detection algorithms. The authors classify the approaches into four categories: two statistical and two deterministic. They propose new quantitative and qualitative metrics to evaluate these algorithms on a benchmark of indoor and outdoor video sequences. The paper also discusses the challenges of moving shadow detection, including the misclassification of shadow points as foreground or background, and the difficulties in distinguishing shadows from objects. The authors compare four representative algorithms: a statistical nonparametric (SNP) approach, a statistical parametric (SP) approach, a deterministic nonmodel-based (DNM1) approach, and a deterministic nonmodel-based (DNM2) approach. The evaluation metrics include detection rate, discrimination rate, robustness to noise, flexibility to shadow strength, and scene independence. The results show that the DNM1 approach performs the most consistently across different sequences, while the DNM2 approach is more effective in specific scenarios. The paper concludes that for general-purpose shadow detection, a pixel-based deterministic nonmodel-based approach (DNM1) provides the best results, while more assumptions are needed for efficient detection in specific environments. The study highlights the importance of considering both statistical and deterministic approaches in shadow detection and emphasizes the need for further research to improve the accuracy and robustness of shadow detection algorithms.This paper presents a comprehensive survey and empirical evaluation of moving shadow detection algorithms. The authors classify the approaches into four categories: two statistical and two deterministic. They propose new quantitative and qualitative metrics to evaluate these algorithms on a benchmark of indoor and outdoor video sequences. The paper also discusses the challenges of moving shadow detection, including the misclassification of shadow points as foreground or background, and the difficulties in distinguishing shadows from objects. The authors compare four representative algorithms: a statistical nonparametric (SNP) approach, a statistical parametric (SP) approach, a deterministic nonmodel-based (DNM1) approach, and a deterministic nonmodel-based (DNM2) approach. The evaluation metrics include detection rate, discrimination rate, robustness to noise, flexibility to shadow strength, and scene independence. The results show that the DNM1 approach performs the most consistently across different sequences, while the DNM2 approach is more effective in specific scenarios. The paper concludes that for general-purpose shadow detection, a pixel-based deterministic nonmodel-based approach (DNM1) provides the best results, while more assumptions are needed for efficient detection in specific environments. The study highlights the importance of considering both statistical and deterministic approaches in shadow detection and emphasizes the need for further research to improve the accuracy and robustness of shadow detection algorithms.