Autonomous Object Detection and Identification

A state of the art approach to computer vision. A new technique is presented termed Edge Flow, set to change the way object detection is thought about.

The main concept was tested on a UAV platform, and the applications for Edge Flow in terms of object detection and identification is extensive.

 

With the ability to detect and differentiate static and moving objects in a scene, the system can be applied to any application which requires the computer system to detect and identify objects.

 

Dangerous or anomalous objects

In scenarios where anomalous objects can be left, or should not be present, the algorithm can detect each object in a scene and identify objects that are abnormal or should not be present. Above we see the detection results on the right. Both the car and bike are detected as the moving object, the road markings as another type of object, the cracks in the road differentiated and the white object on the side of the road is also separated out. Furthermore, even the patches of disturbed earth in the ground are detected as separate objects.

Crop or vegetation analysis

A common problem for farmers is assessing crops in real-time such that bad plants or diseases can be detected early. Because of the capability of the algorithm to differentiate small texture changes (when set up correctly) crops that are performing differently can be quickly identified. Whilst the detection image (bottom right) looks busy and uninformative - the algorithm has actually detected the different variations in the grass topology and separated out the plants that are different as well as the person. The raw output isn't so useful but the data separation and analysis that can be performed with this information is incredibly useful.

Safety and congestion analysis for transport

Areas such as train platforms where there are safety concerns (alighting, boarding and approaching trains) the work can be used to identify passengers in or approaching a danger area in real-time - enabling staff or operators to act accordingly. Furthermore, the work can be used to analyse each object on a carriage assessing how busy each coach is (how much luggage, how many passengers etc). This can be useful information for train operators during peak hours of service.

 

Above is a small selection of the capabilities of the Edge Flow algorithm with respect to Autonomous Object Detection and Identification. For further details or questions please feel free to email the researcher on g.morris2@lancaster.ac.uk