Fall detection systems are essential in health care, sports, and surveillance but often rely on costly equipment like high-resolution cameras or specialized sensors, limiting their scalability and accessibility. This study systematically evaluates existing machine learning (ML) and deep learning (DL) architectures, revealing that traditional methods, such as biomechanical modeling and sensor-based techniques struggle with challenges like occlusions, rapid posture changes, and environmental noise, which reduce their reliability (Bhat et al. in Sensors (Basel) 20:5356, 2020; Cakmak et al. in Late Fusion of Machine Learning Models Using Passively Captured Interpersonal Social Interactions and Motion from Smartphones Predicts Decompensation in Heart Failure, 2021; Mubashir et al. in Neurocomputing 100:144–152, 2013; Pannurat et al. in Sensors 14:12900–12936, 2014). To address these issues, we propose a novel fall detection model that uses local depth pixel changes for body position identification and movement recognition, eliminating the need for skeletal tracking or specialized hardware (Planinc and Kampel in Personal and Ubiquitous Computing, 2012; Rougier et al. in IEEE Trans Circuits and Systems for Video Technology 21:611–622, 2011). The model incorporates hybrid architectures, including CNN-LSTM and two-stream CNNs (Feichtenhofer et al. in Convolutional Two-Stream Network Fusion for Video Action Recognition, 2016; Gadzicki et al. in 2020 IEEE 23rd International Conference on Information Fusion (FUSION), pp. 1–6, 2020), which enhance spatiotemporal feature extraction for robust detection. Tested on the IXMAS dataset (Blank et al. in The Tenth IEEE International Conference on Computer Vision (ICCV’05), pp. 1395–1402, 2005), our model achieved an accuracy of 95.45%, precision of 92.75%, recall of 95.63%, and an F1-score of 0.81—showing a 15% improvement over traditional methods. This scalable, cost-effective solution offers enhanced reliability and generalizability, making it highly suitable for real-world applications. The results highlight the potential for integrating advanced deep learning techniques into fall detection systems to overcome current limitations and provide significant societal benefits, especially for the elderly and healthcare industries (Kapinski et al. in Late Fusion of Deep Learning and Hand-Crafted Features for Achilles Tendon Healing Monitoring, 2019; Stone and Skubic in IEEE Trans Biomed Eng 60:2925–2932, 2013).