Posts Tagged: deep learning

Home » deep learning

Smart Eyes of Smart Cities

Can something as common as a camera be transformed into a complex mobile monitoring system for advanced analysis of traffic? Can we distinguish anomalous and aggressive behavior of drivers, acquire full trajectories of vehicles or even detect traffic accidents before they happen? These are some of the questions that the newest edition of a smart city magazine called D!CITY tries to answer, which included also two smart city solutions using cameras: DataFromSky and ParkingDetection and also smart city platform of our partner OMS Lighting.

In the past, the main task for cameras was to record stories, which could be, unfortunately, interpreted only by humans. Nowadays, thanks to the increasing availability of high-quality cameras and mainly due to the immense technological advancement in the field of artificial intelligence and machine learning, the camera has become a universal sensor with application in many fields of the smart city concept, as they can serve as “the smart eyes of smart cities”. This marks the beginning of a new era, where image can be processed also by machines and where camera represents a rich source of data at even very low costs. What are the main areas of traffic analysis, where a camera already proves itself in action?

In times where individual transport increases faster than ever before, traffic congestions or even collapses can occur on a daily basis. Fortunately, there are systems which can help to achieve more fluent traffic even without expensive infrastructure changes, you just need a certain bird’s eye view. If we give camera “wings”, for example by attaching it to an unmanned aerial vehicle, we get a monitoring system for advanced traffic analysis which is not only very accurate but also at an affordable price. This is exactly what our project DataFromSky deals with – as there are several research projects that require just this type of traffic analysis. The system can detect vehicles and other participants of the traffic flow, which includes not only their trajectories and speed but also detection of possibly dangerous behavior with deep-learning and artificial intelligence. This can even detect traffic accidents before they actually happen, which can provide the traffic engineers with very valuable data about causes of traffic accidents within the tested area in order to look for ways how to avoid them in the future with convenient preventive measures.

Cameras are becoming a prominent technology for traffic analysis also at so-called “adaptive intersections”, as they can accurately detect not only vehicles, but also pedestrians and cyclists, which also play an important role in the given traffic flow. With smart traffic lights control, based on data from a network of sensors, the whole system becomes way more fluent and flexible when it comes to reaction to rapid changes within the traffic, making it also a lot safer. Another use of cameras lies in the field of smart lighting, as the camera can collect visual data and use to analyze public spaces, monitor traffic or detect anomalous behavior. This gives us many possibilities of designing smart streets with the use of a smart camera system, using something as common as public lighting.

Traffic congestion can be also caused by the situation in parking, especially in situations when drivers are trying to find an empty parking space in an infrastructure, which is already overloaded. One of the possible solutions is to use current parking spots more effectively or even navigate drivers to parking lots with available places to park their car. This is the main area of focus of the system called ParkingDetection. As far as technical and economic aspects are concerned, a camera represents a suitable sensorial solution for monitoring of even tens or hundreds of parking spots, as even very large parking lots can be covered by a single smart camera at very low costs, while magnetic sensors would be extremely expensive and even less accurate. If we also add an ALPR camera for number plate reading into the system, we get a significant number of potential technical solutions for residential parking, reservation of individual parking spots or even increased safety within the parking lot.

Are we on the brink of an age of cameras? In the not-so-distant past, autonomous vehicles, or using drones for traffic analysis would sound like science fiction. Nowadays, these technologies are already becoming a reality, also thanks to the advancement in cameras and artificial intelligence. It is becoming clear that cameras will play an important role in the future of smart cities.

The whole magazine can be found at this link.

Read more

Release of DataFromSky LIGHT platform!

Become a traffic expert and send you data for DFS Light analysis right now and you will get some Free Credits to the start!

The DataFromSky Light is a fully automatic online service platform that detects, classifies and tracks all traffic objects in any video source using artificial intelligence, deep learning methods, and the most advanced image processing algorithms trained on millions of traffic situations. We transformed our experience from aerial videos and created the ultimate solution for video traffic analysis for low altitude cameras which is capable of learning from its mistakes. Try and see for free!


DFS LIGHT platform is here

Mining the traffic data was never easier:

1) visit our newly released web platform light.datafromsky.com and send your first video for automated analysis or create a User Account via an online sign-up form
2) upload your video, take a short break and wait for the results
3) you will receive an email with a link to download the DFS Light analysis or you can download it directly from your User account
4) analyze your data with our super tool DataFromSky Viewer Light… visit our Help section that will guide you through on how to use the Viewer and its functions

Why should you use DFS LIGHT? Countless benefits…

  • comprehensive analysis – counting, OD-matrix, turn counts, statistics, trajectories, exports and visualizations and much more just in a few clicks
  • supports up to 16 object classes – from car, truck, van up to tuk-tuk
  • valuable video data – analyze every single millisecond of your video
  • supports up to 8k videos – you won’t miss anything, detection of even the tiniest objects
  • data available in offline mode – download the results and SW to your PC and you will never lose the data
  • no need of expensive hardware – use any type of camera
  • Not yet convinced?
    Try it on your own video, it is easy and you get a short sample analysis for free from each of your video!

    DataFromSKy Light platform is a global platform, however, for the use of the service in Nordic countries or Benelux, please contact our DataFromSky Partners:

    COWI A/S


    COWI

    COWI is a leading consulting group with a 360° approach. The Danish company has a long-proven record of success, serving customers from the Nordic countries and the whole world in the field of civil engineering, including traffic analysis. DataFromSky cooperates with COWI on Nordic markets, including Denmark, Norway, Sweden, Finland and Iceland. For more info about DataFromSKy LIGHT there, contact Mr. Jonas Olesen: jool@cowi.com


    ROELOFS

    Roelofs Groep is a company with over 50-year history in the Dutch infrastructure. They focus on sustainable innovation of infrastructure, belonging into leading infrastructure specialist in the Netherlands. They are involved in various projects in the fields of energy & environment, resources, ecology, mobility, sewage and infrastructure engineering. DataFromSky cooperates with Roelofs in the region of BENELUX – Belgium, Netherlands and Luxembourg. For more info about DataFromSky LIGHT there, contact Mr. T. Adema: t.adema@roelofsgroep.nl




    DataFromSKy LIGHT ≠ DataFromSky Aerial! Do you need super accurate traffic data of an aerial video? Speed measurement [km/h], acceleration data, heatmaps, time gaps, safety analysis are parameters of your interest? These values are not reachable with DataFromSky Light. Contact us for more information!

    Read more

    NVIDIA wrote article about DataFromSky!

    We are happy to share the excitement from the article which NVIDIA, well-known GPU’s provider and AI focused tech company, published about DataFromSky – our AI in the Sky!

    As the article further explains, we train our deep neural networks using thousands of hours of video footage from around the globe, shot in various weather conditions. The training takes place on NVIDIA GPUs using Caffe and TensorFlow. These specialized neural networks can then recognize objects of interest and continually track them in video footage. The data captured via this AI process is used in numerous research projects, enabling deeper analysis of object interaction and new behavioral models of drivers in specific traffic situations. Ultimately, this kind of data will also be crucial for the development of autonomous vehicles!

    We are also carrying out extensive optimizations of the whole architecture, moving to NVIDIA Jetson AGX Xavier, in order to be able to analyze data in real time. Our current work focus is also on the development of an online monitoring drone, which will be able to evaluate image data in real-time, directly during its flight. This type of drone can be used, for example, as a substitute for a police helicopter, or can operate as a mobile surveillance system for personal and property protection.

    Read the full article on NVIDIA blog at this link.

    We are also a member of the NVIDIA Inception program – virtual accelerator program led by NVIDIA nurtures dedicated and exceptional technological companies, who are revolutionizing industries with advances in AI and data science and help them in all stages of product development, prototyping, and deployment.

    Read more

    DataFromSky supported by NVIDIA

    We are proud to announce, that we become a member of NVIDIA Inception Program! This virtual accelerator program led by NVIDIA nurtures dedicated and exceptional startup companies, who are revolutionizing industries with advances in AI and data science, and help them in all stages of product development, prototyping, and deployment. Using NVIDIA Pascal GPUs and Caffe’s deep learning framework, DataFromSky is able to do the automatic extraction of time-space-trajectories of moving objects in aerial video data captured by drones and/or balloons. The overall system is divided into three main parts: preprocessing, vehicle detection, and tracking. In the preprocessing step, the acquired image gets undistorted and geo-registered against a user-selected reference frame. The current solution for detection and tracking utilizes machine powered by NVidia GTX 1080 cards.

    We designed a highly optimized multi-layer detector and trained it on a hand annotated training dataset with 10.2M positive and 10.2M negative samples per class. The detector is capable of learning from its errors and improving over time. Due to the outstanding performance of detection and tracking algorithms, DataFromSky can provide high-quality data and guarantee a hit rate of over 96%. See the raw output from DFS deep neural networks below:

    Read more