There are a variety of people counting technologies, ranging from simple ones with limited capabilities such as break beams and special people counting cameras to more comprehensive with more advanced analytics features such as SaaS AI solutions.
This type refers to stereo vision cameras with a separate image sensor designed for the exclusive purpose of counting people entering or exiting a particular pre-defined space. Inspired by the binocular vision in humans, such cameras usually have two or more horizontally-placed sensors that observes the objects from two different point of views, compare those differing images and then create 3D geometric information of both the environment and objects such as humans, shopping carts and baby strolls.
1. Depth perception:
Special people counting cameras’ capability to perceive depth enables them to distinguish humans from other objects such as shopping carts with a high accuracy rate. Furthermore, with specific calibration of lighting in the environment, they can eliminate false positives due to shadows to great extent; boosting accuracy rate.
2. Low Bandwidth consumption:
Since most of the computations are happening on the camera itself such a system has a very little data footprint. More often than not videos are not recorded at all so only raw numbers are transferred through the internet for further aggregation and display.
1. Only one piece of the puzzle, not a comprehensive solution:
As their name implies, special counting cameras are limited to performing one task well, namely the people counting and they are not fit for deriving value from retail analytics. For example, these cameras cannot enable a store to optimize its store layout, provide better products or improve customer experience in the store because they do not collect or analyze data about customer demographics or behavior. Neither can these cameras be configured to detect security threats and send alarms.
2. Limited field of view:
Furthermore, these cameras are designed to look down on the floor; observing only the top of the objects from a narrow angle with limited room for tilting and adjusting. This increases the complexity of installation and the cost of covering wide areas. Some of the best people counting cameras can cover up to 40 square ft (3.7m) if installed at the height of 58-60 ft but will cover only 4 square ft (0.3m) if installed at 11ft.
3. Maintenance costs:
From cleaning of the external components such as image sensor and lens to the removal of debris and dust and to general wear & tear, special people counting cameras impose high maintenance costs on businesses. One example is the repair of lenses, the most important part of special people counting cameras. The repair of a 2000$ lens can reportedly cost 200-225$.
4. High cost of purchase and installation:
Since special people counting cameras perform one specific function, they need to be purchased and placed separately from security cameras so they result in extra cost for businesses. Furthermore, these cameras are expensive because of a variety of factors such as special hardware components for stereo calculations and 3D vision.
Expensive special people counting cameras became highly popular because they were doing the one task exceptionally well - people counting. However, recent developments in the computer vision field such as access to vast amounts of data and sufficient computing power required for AI provides businesses with more opportunities to improve accuracy of people counting and reap the full benefits of retail analytics.
SaaS Computer vision systems do not require installation of separate cameras for people counting as they use conventional CCTV cameras as a source of information. In this system, video/image frames recorded by IP cameras are either sent to the Cloud or to in-house server for analysis.
After videos are processed by a computer vision model humans are classified with the context of the scenes and their behavior so more advanced analytics could be performed.
1. Enhanced customer experience:
Computer vision powering conventional cameras can help measure customer dwelling time around particular areas of a store so that the business can improve the customer experience. This may include taking actions such as putting drinks on lower shelves to make them easily accessible to customers.(If customers were having a hard time reaching out to top shelves and were constantly asking for help). Furthermore, analytics insights from computer vision models can show the specific places that need staff assistance in real-time and help direct available staff to that area
2. More targeted marketing campaigns:
Computer vision systems can help determine which demographic segment is more interested in which product category based on the dwelling time around certain aisles, on measurement of customer engagement and the tracking of individuals. For example, a computer vision system can help detect which age or gender group is more interested in a newly-featured vegan chocolate bar. If the system reveals that men aged between 20-30 dwell the most around these candy bars, the store can tailor its marketing campaigns based on this insight, select the most appropriate marketing channel to reach this demographic group and craft its message accordingly. This will in turn lead to more sales.
3. More accurate people counting:
Contrary to special people counting cameras that are not capable of distinguishing between different groups of people such as employees and customers, computer vision systems can be trained to detect the employees who enter or leave the store and remove them from the counting and analysis process so that more accurate results are provided. For example, a consultant may frequently exit and re-enter a certain area to help customers. Computer vision systems can detect the consultant based on his/her distinguishing features such as a name tag or outfit/uniform and eliminate it from people counting.
4. Analyze Employee performance:
AI's capability to single out, detect, track and analyze employees and customers across the store can offer valuable insights about employee performance, how to improve it and help drive up key performance metrics. For example, retailers can measure the number of interactions each employee has, learn about which of these interactions led to a purchase and more importantly, understand how to enable each employee to perform his/her best; including pairing them with customers belonging to a particular demographic group or locating them in a specific area in store.
5. Uncover and Eliminate inefficiencies:
Another key benefit of AI-powered people counting cameras is that you can uncover inefficiencies that can be eliminated and as a result, reduce costs and improve employee performance. For instance, insights from AI-powered cameras may reveal that your employees waste substantial time with tasks such as restocking and fixing planograms; leaving less for employees to help customers or even ignore them altogether. This can help you devise new ways to cut down on time wasted on such tasks and boost their performance and productivity.
6. Cost:
Unlike specialized people counting cameras that have a high price tag and also come with high maintenance and installation costs, AI-powered cameras are a cost-effective solution that does not require heavy investment. First of all, AI-powered cameras are not installed separately; instead, AI is connected to conventional security cameras that are already installed onsite. There is no need to send an installation team, and there are no extra costs for acquiring new hardware. Secondly, since they do not require the installation of separate cameras, the problem of double-maintenance-cost is eliminated. Third cost-related benefit is that implementation of AI-powered solutions requires fewer numbers of cameras, contrary to specialized-people counting cameras needed to cover the same retail space. In a retail space with 4-way crossing, for instance, one A.I powered camera would be sufficient to cover all areas of interest. Specialized people counting cameras, on the other hand, would require 4 separate cameras to cover every crossing separately.
The main disadvantage of A.I-powered cameras is it takes a significant amount of time to fine-tune the model, a useful technique to ensure that a previously-built AI algorithm can generalize into the new, unknown retail environment and achieves the desired level of accuracy and robustness in different tasks such as people counting, dwell time measurement, employee & customer classification and retail analytics.
Consider a child who spent his/her entire childhood around a lake that only had white swans. When this child moves to a new place and comes across black swans, he/she would miserably fail to classify black swans as ‘swans’ because his knowledge of swans was limited to those with white feathers.
AI algorithms are similar to humans: When you decide to use AI-powered cameras, you need to account for the differences in the new retail space such as differences in lighting, camera replacement, the overall layout of the store and the different ways in how customers interact with products. These differences can result in a lower accuracy rate in different tasks such as dwell time measurement, double counting reduction and retail analytics Fine-tuning adjusts the AI algorithm for the new retail environments and helps you achieve the target accuracy levels and ensure the robustness of the AI-powered cameras.
For example, if the existing AI model was trained with data of employees wearing uniforms but the employees in the new retail store are not doing so, the model needs to be adjusted to perform with the same accuracy in detecting and classifying employees in the new retail space. Other differences to account for may include the more frequent occlusion of people due to narrow pathways in the retail space, the different position of camera or the effect of lightening. Considering the number of trial-errors that may be involved in the fine-tuning process, the configuration of the A.I model that powers security cameras may take up to 3-4 weeks. While fine-tuning is a time-consuming task, it does not degrade the value you can extract from using AI powered cameras if you implement fine-tuning effectively.
In fact, fine-tuning can provide you substantial benefits such as:
Saving you from the costly investment in building a custom AI solution from scratch. Fine-tuning of AI is significantly cheaper and faster compared to creating a new one. You can adopt an existing AI solution and tailor it to your particular retail environment.
Fine-tuning may be performed on a regular basis and happen in the background allowing you to experiment with new strategies, launch new marketing initiatives in an ever-changing environment and accurately measure the impact of every change.
Fine-tuning can play a vital role in helping you create a unique in-store customer experience and help you distinguish your business from your competitors. The benefits of a properly fine-tuned AI-powered camera will far outweigh its one drawback: the amount of time it takes.
When it comes to choosing the right people counting solution, there is no one-size-fits-all answer due to two obvious reasons:
Firstly, each business has different goals by implementing people counting technology:
While one retail business may be satisfied with learning about the number of visitors per hour, another retail store’s goal may be to better understand customer behaviour, improve customer experience and measure customer engagement with certain areas in its store.
Secondly, each environment where people counting technology is deployed differs widely due to factors such as lighting, occlusion, position of cameras, variety and number of unknown objects. All these factors may have a significant impact on the accuracy rate and performance.
If you have a limited number of exits and entrances and you are interested in collecting foot traffic data, special people counting cameras is the way to go. You should keep in mind the high maintenance costs, high price tag and the cumbersome process of synchronizing separate special counting cameras to produce accurate results on a high scale.
If you want to utilize people counting technology to drive your business forward, AI will be a better choice compared to special people counting cameras that merely observe the number of visitors. Computer vision extracts deeper insights of customer behaviors, learns more about customer demographics, achieves higher accuracy rate at measuring foot traffic and improves customer experience. You should also choose Computer Vision systems over other options to if you need a versatile solution and quick adaptation to new challenges.
Traces A.I helps you to decipher the meaning and insights behind each raw number collected from the people counting system, whether it is just the number of visitors or employee performance.
Curious about how Traces AI enable you:
Contact us at: info@traces.ai
.jpg)
