Embedded computer vision demos



Traffic monitoring using linesensor

The use of Wireless Multimedia Sensor Networks (WMSNs) in Intelligent Transportation Systems (ITS) can offer cost effective solutions for gathering data on urban traffic, vehicle velocity, parking, etc. 
These applications demand real-time image acquisition and computational intensive processing in order to collect relevant information. We present our hardware and software solution for detecting moving objects and evaluating their velocity based on line sensor technology which guarantees fast processing, low storage and bandwidth requirements.



 Gaussian Mixture Background Modelling Optimisation for Micro-controllers

This work investigates sub-integer implementations of the adaptive Gaussian mixture model (GMM) for background/foreground segmentation to allow the deployment of the method on low cost/low power processors that lack Floating Point Unit (FPU). We propose two novel computer integer arithmetics to update Gaussian parameters. Specifically, the mean value and the variance of each Gaussian are updated by a redefined and generalised round operation that emulates the original updating rules for a large set of learning rates.

Weights are represented by counters that are updated following stochastic rules to allow a wider range of learning rates and the weight trend is approximated by a line or a staircase. We demonstrate that the memory footprint and computational cost of GMM are significantly reduced, without significantly affecting the performance of background/foreground segmentation.



Parking spaces occupancy algorithm

Wireless Sensor Network (WSN) has been adopted during the years for a large set of applications, such as environmental monitoring, industrial automation, process control, and, more recently, for video surveillance and multimedia streaming applications.
Surveillance applications like traffic monitoring, vehicle parking control, and intrusion detection involve monitoring of the environment in order to detect and interpret the activities of relevant objects and their behavior. These applications demand real-time images from the scene in order to collect relevant information. Wireless Multimedia Sensor Network (WMSN) provides a low-cost and flexible solution for distributed video surveillance based on low-power autonomous smart camera nodes. Image processing in the WMSNs scenario is very challenging due the necessity to store and process huge amount of visual data on energy and resource constrained sensor nodes. To deal with this problem, many researchers have mainly proposed distributed in-network processing and low-complexity computer vision techniques to be executed on-board.
We focus on the development of on-board image processing techniques for detecting the occupancy status of parking spaces. The developed techniques are presented as an effective solution for vehicle parking lot monitoring applications in the domain of Intelligent Transportation Systems (ITS). Performance results show as the developed mage processing techniques are able to reach 99.92% sensitivity and 95.59% specificity in detecting the parking spaces occupancy status. Moreover, in the chapter a full implementation of the developed algorithms in embedded devices is presented and their overall performance evaluated in terms of execution time and memory occupancy.



fot architecture