SkyVision: Customer Flow Misreporting Reduced by −75%, Halving Recertification Effort

DaoAI SkyVision 0-code video surveillance AI platform, integrating DaoAI World world model's semantic understanding, successfully reduced misreporting in customer flow counting and heatmap analysis by −75% and decreased related data recertification effort by −60% for a large retail chain, significantly enhancing data analysis reliability and operational efficiency. In smart surveillance, accurate customer flow data is crucial for precise operations, resource optimization, and marketing effectiveness evaluation in retail, venues, and transportation hubs. However, traditional video surveillance systems often face high false positive rates in complex environments, particularly in customer flow counting and heatmap generation, where misreporting not only compromises data accuracy but also leads to extensive manual recertification.

DaoAI SkyVision 0-code video surveillance AI platform (on-site hourly training of proprietary models, behavior/event recognition, real-time alerts from edge devices, 100% on-premise data, DaoAI World world model for semantic understanding) has successfully integrated large model technology to effectively address the long-standing false positive challenge in traditional customer flow counting and heatmap analysis. It reduced the misreporting rate for customer flow statistics in large commercial complexes from an industry average of 15-20% to <4%, while decreasing manual recertification effort by −60%. In the smart surveillance industry, precise customer flow statistics and heatmap analysis are critical pillars for various domains such as business operations, urban management, and security. Whether it's shopping malls, transportation hubs, or exhibition venues, real-time perception of human traffic and deep insights into behavioral trajectories directly impact the efficiency and accuracy of resource allocation, marketing strategy formulation, and security alert responses. However, most video surveillance systems on the market currently struggle with high false positive rates in complex environments, particularly in customer flow counting and heatmap generation, where counting discrepancies and heatmap distortions are common, severely affecting the practical value of the data. The advent of DaoAI SkyVision aims to fill this gap by providing a more reliable and accurate customer flow analysis solution through advanced AI technology.

Pain Points: Why This Hurdle Is So Difficult to Overcome

In customer flow counting and heatmap analysis scenarios, the pain points of traditional solutions primarily revolve around three dimensions: data distortion and decision bias due to high false positive rates, immense manual recertification workload, and the complexity of deployment and maintenance.

Firstly, traditional customer flow counting systems based on background subtraction or simple feature matching generally suffer from high false positive rates in complex and dynamic environments, often averaging 15-20%. For example, in shopping malls, trolleys, strollers, large luggage, or even light reflections can be mistakenly identified as 'people,' leading to inflated customer flow data. In densely populated areas, target occlusion and crossing movements are common, easily resulting in undercounting or overcounting. Such inaccurate data directly impacts business operational decisions, such as evaluating promotional effectiveness, optimizing staff scheduling, and analyzing area popularity, potentially leading to incorrect resource allocation and marketing strategies. Secondly, to ensure data usability, a high false positive rate inevitably leads to a significant amount of post-processing manual review and correction. For a retail chain with hundreds of surveillance points, the vast number of abnormal alerts and inaccurate data generated daily requires a dedicated team to spend considerable time on video playback, manual counting, and data correction. This not only incurs substantial labor costs but also delays the real-time availability of data analysis, making timely decisions difficult. Finally, traditional solutions often require time-consuming and labor-intensive parameter adjustments or even model retraining when facing new scenarios, business models, or environmental changes. They lack flexibility and generalization capabilities, resulting in high deployment and maintenance costs and difficulty in rapidly adapting to business needs.

Technical Principles

The reason why DaoAI SkyVision platform can significantly reduce false positive rates lies in its integration of advanced deep learning algorithms, few-shot learning capabilities, and the semantic understanding provided by the DaoAI World world model. Unlike traditional algorithms that rely on fixed rules, SkyVision employs a Transformer-based vision foundation model capable of learning more robust and generalized feature representations from massive video data. By introducing self-supervised and contrastive learning mechanisms, the model can precisely recognize human forms, postures, and movement patterns without requiring extensive annotated data. When faced with complex scenarios, the platform leverages its powerful contextual understanding, such as distinguishing pedestrians from trolleys or filtering out light reflections, thereby substantially reducing false positives. Furthermore, DaoAI SkyVision's few-shot learning capability is crucial. During on-site deployment, the platform supports hourly training of proprietary models, requiring only a small number (1-20) of real-world 'positive' or 'negative' sample images to quickly adapt to specific environments, such as identifying employees in specific uniforms or distinguishing particular objects. This effectively addresses the generalization limitations of general models in specific scenarios, enabling rapid local iteration and optimization of the model, ensuring high accuracy.

Compared to traditional computer vision methods based on background subtraction or Kalman filtering, DaoAI SkyVision's advantages are evident in several aspects: Firstly, traditional methods are sensitive to noise and prone to numerous false positives under complex backgrounds and varying lighting, whereas SkyVision's deep learning-based feature extraction effectively filters out such interference. Secondly, traditional methods suffer from significant accuracy drops in counting when targets are occluded or overlap, while SkyVision maintains high counting accuracy in complex scenarios through more refined pose estimation and multi-object tracking algorithms. Finally, traditional methods typically require extensive manual parameter tuning or expert experience, leading to long deployment cycles and high maintenance costs. In contrast, SkyVision's 0-code, few-shot training features enable even non-professionals to quickly get started, achieving hourly model training and deployment on-site, significantly lowering technical barriers and operational costs. Combined with real-time alerts from edge devices and a 100% on-premise data policy, it ensures data security and immediate response.

Typical Application Scenarios

• **Retail Store Customer Flow Statistics and Conversion Rate Analysis:** In large supermarkets or brand stores, SkyVision can real-time count customer entries, dwell times in specific areas, and calculate conversion rates by combining with POS data. The challenge lies in distinguishing between customers and staff, excluding non-shopping behaviors, and accurately counting during peak promotional periods. SkyVision ensures data purity through behavior recognition and semantic understanding.
• **Commercial Complex Heatmaps and Traffic Flow Optimization:** Visualizing pedestrian density and movement trajectories across different floors and areas of a shopping mall to generate heatmaps. Challenges include multi-camera collaboration, continuous target tracking in large scenes, and privacy protection. SkyVision offers multi-camera fusion technology and supports anonymization, assisting malls in optimizing lease arrangements and spatial layouts.
• **Public Venue Congestion Warning and Emergency Management:** In public places such as subway stations, train stations, and large exhibition halls, real-time monitoring of crowd density, automatically triggering congestion alerts when predefined thresholds are met. The difficulty lies in accurate counting during extreme congestion and rapid response. SkyVision's edge device real-time alert capability ensures alert messages are delivered in seconds, providing valuable time for emergency management.
• **Exhibition Visitor Behavior Analysis and Booth Optimization:** Analyzing visitor numbers, dwell times, and areas of interest at different booths during exhibitions to evaluate booth design and product display effectiveness. Challenges include rapid adaptation to temporary setups and precise data collection for short-term events. SkyVision's 0-code, hourly training capability allows it to quickly adapt to exhibition scenarios, providing customized analytical reports.

Case Study

A leading chain supermarket group, with over 500 stores, has long been troubled by inaccurate customer flow statistics and high manual recertification costs. Their original system primarily relied on traditional vision algorithms for customer counting, with a false positive rate as high as 18%. Especially during peak hours, shelf replenishment, staff patrols, and customers carrying large items were often mistakenly identified as new customer traffic, leading to severe data distortion. Consequently, the group had to invest significant human resources (approximately 2000 man-days per month) in video playback and data correction, which was not only costly but also greatly reduced the timeliness of customer flow data, preventing effective guidance for store operations. After introducing the DaoAI SkyVision platform, we first conducted on-site hourly model training for their typical store scenarios (entrances, cashier areas, fresh produce sections). By collecting a small number (fewer than 20 per scenario) of real false positive samples (e.g., stock clerks, trolleys) and correct samples, the SkyVision platform completed customized model deployment in just a few hours. After going live, a one-month actual operation verification showed that the customer flow counting false positive rate in the group's stores dropped from the original 18% to <4%, achieving a significant reduction of −77%. More importantly, due to the substantial improvement in data accuracy, the monthly man-hours spent on manual recertification decreased by −60%, greatly alleviating the burden on the store operations team, allowing them to focus more on customer service and sales strategy optimization. Concurrently, combined with the heatmap analysis provided by SkyVision, the group could more accurately identify 'golden zones' and 'cold zones' in their stores, optimizing product displays and promotional activities, thereby improving sales per square meter.

DaoAI SkyVision's 0-code platform, with its on-site hourly training and large model semantic understanding, reduced customer flow misreporting from 18% to <4%, saving −60% manual recertification effort monthly, truly enabling data-driven refined operations.

DaoAI Solutions and Products

The core solution provided by DaoAI to clients is based on the SkyVision 0-code video surveillance AI platform. The platform's key advantage lies in its '0-code' and 'on-site hourly training of proprietary models' capabilities, significantly lowering the barrier to AI application and reducing deployment cycles. In customer flow counting and heatmap scenarios, we first seamlessly integrate with the client's existing surveillance cameras via SkyVision's video access module. Subsequently, using the platform's built-in intelligent analysis algorithms, pedestrians in the video stream are detected, tracked, and counted in real-time. For false positive issues in specific scenarios, such as distinguishing trolleys from pedestrians, the client's on-site operational or technical personnel only need to upload a small number (typically 1-20) of false positive or false negative images as training samples through SkyVision's intuitive interface. The platform can then perform rapid model fine-tuning on local edge devices (supporting 100% on-premise deployment, with data never leaving the premises) for hourly model optimization and iteration. This rapid adaptation capability, combined with the semantic understanding of the DaoAI World world model, enables SkyVision to comprehend more complex scenarios and behaviors, thereby significantly improving recognition accuracy and keeping the false positive rate at a very low level. Concurrently, the platform supports real-time alert pushes, instantly notifying management personnel when abnormal situations are detected or predefined thresholds are reached, ensuring quick response. All data can be stored and processed locally, complying with strict data security and privacy protection requirements.

Through the DaoAI SkyVision platform, clients not only gain highly accurate customer flow statistics and heatmap data but, more importantly, achieve significant improvements in operational efficiency and effective cost control. The reduction in false positive rates directly decreases the workload of manual recertification, freeing up valuable human resources; the timeliness and accuracy of data provide reliable decision-making basis for management, thereby optimizing resource allocation, enhancing customer experience, and ultimately promoting business growth. For instance, with precise heatmaps, a particular shopping mall was able to re-plan its promotional areas, leading to a 15% increase in sales for specific zones.

This article was generated by AI. Customer cases are simulated scenarios based on real product capabilities and figures are illustrative; see product pages for official benchmarks.