e-con Systems offers a multi-camera system for NVIDIA Jetson AGX Xavier Developer kit and NVIDIA Jetson TX2 Developer kit. Called SurveilsSQUAD, this multi-camera system is based on the SONY STARVIS IMX290 image sensor. With four SONY Starvis IMX290 – 2MP sensors working in synchronization, SurveillsQUAD provides seamless embedded vision with 360-degree field of view suitable for parking lot management. Connected to NVIDIA on-board Jetson Xavier Image Signal Processor (ISP), it performs various auto-corrective functions that give you significant improvements in image quality.
To demonstrate SurveillsQUAD in action, e-con Systems has released a application example. This Parking Lot Management application leverages NVIDIA’s DeepStream SDK 4.0—a DIY comprehensive analytics toolkit for image and video-based applications. The NVIDIA Xavier platform enables simultaneous multi-stream processing. NVIDIA’s Deepstream framework offers plugins for image processing, computer vision and deep learning. These can be connected and developed into a stream processing pipeline. The application uses the deep learning packages and image processing algorithms from the SDK for intelligent vehicle tracking and parking lot management.
DeepStream SDK 4.0 and Visionworks APIs (for dense optical flow estimation) detects the motion of vehicles in your parking lot. Hence, you can understand their direction and report on whether they are entering or exiting the premises. All the pixel-related information about vehicles are then presented to you as data-rich actionable insights.
Key components for Parking Lot Management
Nvinfer: Nvinfer is the primary and secondary infer engine used for car detection and classification. The primary engine detects the presence of a car in the parking lot; the secondary engine identifies the car color and type of model.
gst-dsexample: gst-dsexample is a customized sample application for detecting the movement and direction of the recognized cars. Visionworks SDK and OpenCV, a motion estimation algorithm, analyzes any motion in the scene, as well as the ROI regions of detected cars. Hence, it becomes easy to calculate the direction of the respective vehicles.
NvOSD: After updating the direction data from the dsexample in the metadata of the respective frames, NvOSD plugin writes the respective content on the scene.
The DeepStream SDK 4.0 comes prepackaged with sample applications. It provides real-time edge analytics to capture deep-dive insights. The SDK provides easy integration with cloud service providers, containerized deployment and support for Jetson platform.
The diagram below shows how SurveilsQUAD and DeepStream SDK 4.0 work in tandem.
e-con Systems | www.e-consystems.com