Currently, many Indian enterprises and businesses have evolved to create data-driven cultures internally in their organisations. The primary driver is the adoption of new technologies like machine learning and AI-driven analytics that help them make faster decisions. However, a large number of the current analytics products in the market are not relevant to offline businesses like colleges, brick and mortar stores, etc., where they have to rely mostly on assumptions. India provides an opportunity as the largest untapped source of visual data that can be converted to insights and actions through the use of Edge IoT and AI Vision.
With a vision of building a general-purpose platform to make any “visual sense,” Bangalore-based Wesense.ai is utilising artificial intelligence and machine learning to fulfil the dream. After running pilots across various use cases using CCTVs in hospitals, offices, schools, restaurants, etc., the founders found an excellent product-market fit for the retail sector where extracting visual insights from a brick and mortar store had a significant impact for the businesses. Wesense raised an angel round of 1.3 crores fund last year from serial entrepreneurs.
CCTV Intelligence platform, Wesense.ai was founded by Sankara Srinivasan Aiyyathurai, Alok Mishra, Soumya Ranjan Mohapatra, and Debesh Kuanr in May 2017. To unleash the enormous hidden potential of CCTVs by integrating it with our AI-driven analytics solutions, Wesense.ai was born as a platform to make any CCTV camera smart. Wesense offers not only a platform for capturing data from the brick and mortar stores but also an advanced insights platform built on top of that data
Flagship Product
At Wesense.ai, the flagship vision product for retailers has the following modules
- TrafficSense: Total walk-in count, employee exclusion, total dwelling time, age, gender details of walk-in customers.
- AisleSense: Path taken by in-store customers, density maps
- QueueSense: Optimise the wait time of customers.
- AdSense: Total viewer count and demographics, percentage of total customers viewing the ad, time spent in viewing ads, ad heatmaps.
Use of AI and ML
According to the founders, Wesense has patented a process to communicate between thousands of edge devices and their servers using a low latency WebRTC channel with a pub-sub mechanism as a fallback.
The working model is such that the video frame decoding process uses Intel’s recent edge optimisations. The input data is then passed through multiple inference models on edge using async. The company created an internal model zoo which includes various models for different camera angles and heights for face detection, people detection, age gender classification, face recognition, and so on. The models are trained on cloud GPU instances and are deployed to the edge devices based on the “Senses” enabled for that device.
The output of these models is finally sent to the server and picked up by the server analytics module. For analysis, Wesense uses classical machine learning for time-series predictions and other models mostly based on the client requirements.
Core Technology Stack
According to the founders, the core tech stack can be broadly divided into edge and cloud. The edge architecture is similar to AWS DeepLens, but without the hardware and with support for all kinds of CCTV cameras and works with multiple clouds.
On the Edge side, a deep learning and video processing algorithm known as “Sense” is written in C++. They also have native apps which loads the “Sense” on the edge device and are the agent to communicate with the cloud. The models are trained using TensorFlow on the cloud GPU instances and are deployed to the edge devices based on the “Senses” enabled for that device.
On the Cloud side, Wesense uses a serverless architecture with options of AWS and GCloud. Besides, the company also runs a tuned FB Prophet based system on cloud to predict traffic at the stores.
Hiring Phase
During the hiring process, the company prioritise problem-solving ability, culture fit, and aspiration towards growth in leadership and learning. Also, according to the founders, ‘referrals’ have been a significant tool in hiring, as for them, it addresses the trust factor and expedites the hiring process.
Potential Competitors
Companies like RetailNext, ShopperTrak, and V-Count are a few global players who have deployed AI-driven solutions in the retail space are being considered as potential competitors. Also as a video analytics platform, the company is competing with the camera manufacturers who are currently not in a position to run their software on their competitor’s cameras and in general lack vertical-specific features needed for enterprises.
Roadmap
In the next five years, the company wants to develop more senses and intend to grow horizontally as well into other CCTV use cases like offline advertisements, employee monitoring systems, etc.
