Vision Based Surveys

Vision Based Surveys

Tracsis can provide Vision based Survey (VBS) technology anywhere in the UK and Ireland. With swift deployment, accurate analysis and bespoke reporting we are market leaders.

We use the most advanced digital equipment with machine learning algorithms that allow us to analyse complex mixed traffic situations including simultaneous road, cycle lane and footpath coverage. 3G enabled sensors provide data collection and their connectivity gives us live access to carry out system checks and maintenance.

How does it work?

Our sensors use video analytics in order to understand road scenes. Each sensor has an on-board camera, a processor, and 3G connectivity. The camera takes video continuously, and feeds this through to the processor, where machine learning algorithms are used to extract useful data from the complex video feed. The screen capture image below demonstrates the way the algorithm identifying objects from within the video stream.

Figure 1: Machine Learning Software Output

Once objects have been identified and classified, the video feed is discarded, and the anonymous data about the number and class of road user is sent to our cloud servers. This is made available to clients through web interfaces.

Given the 3G connectivity, we have complete access to all sensors at any time. This enables a full, real-time view of sensor up-time, remote debugging and maintenance, and remote updates. The capabilities of our sensors are constantly improving with plans to add detailed speed tracking and exploring the subdivision of classes, such as splitting pedestrians into joggers, parents with children, and groups.

How accurate is it?

Computer vision technology has been used for data collection for decades. However, it is only with the advent of advanced machine learning algorithms (neural networks) in the past few years that computer vision has started to achieve its full potential.

Our current trials have demonstrated 97%+ accuracy in identifying and differentiating pedestrians and cyclists in environments such as those experienced in the Canal & River Trust project described below. In a trial with TfL, we have demonstrated 93%+ accuracy during a 7am-7pm pilot project, even with large pelotons of 20+ cyclists released from traffic signals simultaneously. Our broader road classification work achieves over 90% accuracy across a full 7-class classification scheme, differentiating traditionally difficult to split classes such as motorcyclists from cyclists, and buses from HGVs.

In short, machine learning has now achieved the levels of accuracy required to run surveys projects using our camera sensor systems.

Technology

We provide an intelligent camera solution, which processes video footage locally on the camera, and only transmits the data required by our clients, rather than the full video stream.

Our technology partners Vivacity Labs have developed machine learning software, which we train to recognise different types of vehicle in the video feed. The tracking software provides a live path and speed of each object within the field of view.

Vehicle Classification

Our software has been designed to be able to classify different road users, due to a deep understanding of local road scenes. Accuracy is typically c. 95%, and the system is extremely flexible in terms of looking at a variety of viewpoints & conditions.

Accessing our Data

Our data is accessible through our online dashboard which is under development as part of project work in Milton Keynes and for the Canal & River Trust. Special features can be added to the dashboard and we are happy to meet with clients to discuss specific requirements.

A prototype of the dashboard is shown below, allowing the user to select a sensor (shown in purple), choose a date, and then display the classified count for that particular day. We can interrogate the data for trends, average and unusual behaviours, and to compare different sites. Any given view of the data can then be downloaded.

We can also provide the full data extract in various output formats on request.

Data Protection

Our system is GDPR compliant. During normal operation, we utilise “privacy by design” principles. As all video is processed into anonymous data and then discarded on the device, there is no risk of sharing personal data.

No personal data leaves our devices under normal operation. Our devices are all secured and encrypted, as are our servers, minimising risk of any intrusion or cyber-security risk.

The full pipeline is shown below:

Installation Design & Project Management

Our knowledgeable team plan and execute Vision Based Surveys (VBS) with meticulous accuracy, considering optimum mounting locations and software programming. Watch our video for an illustration of this in practice.

See Vision Base Surveys in action on these case studies: