Case Study:Counting Rods: It’s Not Always 1-2-3
Rods are thrown on a conveyor belt. The belt stops for a short period and starts rolling again. From the the conveyor belt, the rods fall into a nearby cradle. Rod batches are created by this process. Daily rods continually cycle through this process creating batches. The rods must be counted. Surprisingly this simple task is still performed manually. And by counting each bundle multiple times, the “counter” believes his results are reliable. A steel rod factory approached us to automate this process and make the count absolutely accurate..
The Considered Approaches:
There were several approaches which were considered as a solution.
- Laser Technology/Infrared Sensors: Time-of-flight laser sensors can detect, count, trigger, map, profile, scan, and guide as well as verify levels and distances to practically anything. Basically time-of-flight (TOF) describes a variety of methods that measure the time that it takes for an object, particle or acoustic, electromagnetic or other wave to travel a distance through a medium. A time-of-flight camera (TOF camera) is similarly a range imaging camera system that resolves distance based on the known speed of light, measuring the time-of-flight of a light signal between the camera and the subject for each point of the image. The time-of-flight camera is a class of scannerless LIDAR, in which the entire scene is captured with each laser or light pulse, as opposed to point-by-point with a laser beam such as in a scanning LIDAR systems. As a side note, an infrared based digital counter was also considered. After the connection of the counter module and RC module, when the transmitter is oriented towards the infrared receiver module, If the beam is blocked, the circuit will start counting.
- Tactile Counting: For this approach, self made flap/switch sensors would have to be set up on the conveyor belt. When the rod touches the sensor, a count is made.
Figure 01: (a) Laser (b) Tactile Counting
Although the above stated processes are proven to be accurate, our client was reluctant to use it due to the invasiveness of the process deployment and cost. They needed a pure computer vision way, meaning a count from a live video stream of the rods.
- Computer Vision: Humans use their eyes and brain for seeing and processing the world around them. Computer vision is the science to produce a similar, if not better, capability to a computer. Computer vision is a field that comprises methods for acquiring, analyzing, understanding and processing from a single image or a sequence of images. In general it is concerned with high-dimensional data from the real world in order to produce numerical or symbolic information, e.g., in the forms of decisions. So computer vision involves the development of a theoretical and algorithmic basis to achieve automatic visual understanding to a machine. Why was computer vision such an intriguing prospectus? Because once the software is written to solve a particular problem, it can easily be used an unlimited number of times with no particular additional dependencies or effort. There is no doubt that within 20 years computer vision will become a commodity component of analytics infrastructure and so to the telecommunications infrastructure of the world.
For our challenge, all that was needed to solve this rod counting is a vision sensor. The vision sensor has to be placed where it assures a clear view of the rods and adequate luminosity. Then a rod counting algorithm is produced to finalize the process. It is applied frame by frame to the video stream for the intended result, automated counting.
The Chosen Approach Computer Vision And Its Challenges:
One of the main challenges of computer vision is correctly segmenting objects. Another challenge is the lack of availability of a test case. In our case, it was very hard to differentiate some background noise generated by objects from the test case. There was no single image of just background area, so the foreground area couldn’t be separated from that image. For that reason, a particular region of interest has to be set in the image to resolve this issue:
Figure 02: ROI of images
Videos are also very pulsatory. Think of it this way. It is very difficult for a human being constantly shaking his head to also perform a task while shaking. There must be a clear and steady view to perform a task. For that reason, it was very challenging to factor out these obstacles and construct a robust algorithm to provide a dynamic solution through computer vision.
Figure 03: After processing a sample image
Our client’s goal was at minimum 95% accuracy. We have surpassed that, but we are still not able to achieve 100%. We continue to research a solution to reach this. However, we have been able to ensure 100% accuracy by computer vision when an object is on a conveyer belt that is separate and distinguishable.