Novel Defect Tracking Using DeepSORT, ScaledYOLOv4 and the Clip Model

In real time defect detection scenarios, the problem of double counting defects can generate lots of headaches, mostly because it makes the system more prone to a large number of false-positives. This problem arises from the fact that the system processes 30 frames per second and the target object is moving. Hence, there are multiple frames depicting the same inspected object and thus, a defect can be counted multiple times. This is increasing the false positive rate and furthermore this problem affects the economic yield of the production line and is of course, a problem that needs to be tackled. By using DeepSORT together with ScaledYOLOv4 and the OpenAI clip model, we can track defects and hence, get rid of most problems regarding counting the same defect multiple times. Only process specific adjustments needed. And the tracking is still performed in real time. The code can be found in the following github repository below.