Accurate biomass estimation is critical for optimizing feed management, growth monitoring, and harvest planning in commercial aquafarming operations. Traditional methods involving manual sampling and physical measurements are labor-intensive, stressful to aquatic organisms, and provide only sparse temporal data. This article presents a comprehensive review of computer vision-based approaches for non-invasive underwater biomass estimation in controlled aquaculture environments. We examine the development of intelligent imaging systems designed to operate in challenging submerged conditions, including stereoscopic and monocular camera configurations, specialized lighting systems, and adaptive optical designs. The integration of artificial intelligence, particularly deep learning architectures for object detection, segmentation, and regression-based biomass prediction, has enabled automated analysis of fish size, weight, and population density from continuous video streams. Key challenges include underwater image degradation from turbidity, variable illumination, occlusion in dense populations, and the need for species-specific calibration models. Applications span cage-based marine aquaculture, recirculating aquaculture systems, and offshore farming installations, where real-time biomass monitoring supports precision feeding strategies and early detection of growth anomalies. Future developments will focus on improved generalization across species and environmental conditions, integration with digital twin frameworks for holistic farm management, and deployment of autonomous underwater vehicles for large-scale monitoring. Computer vision-based biomass estimation represents a transformative technology enabling sustainable intensification and data-driven decision-making in modern aquafarming.