A sample project use Apple CoreML & Vision to detect traffic lights
This project doesn't contain a ML model, that means you can't just download and run, you need to find a model or train one by yourself. The One I used for testing is from davidbrai's project [Recognizing Traffic Lights with Deep Learning] (https://github.com/davidbrai/deep-learning-traffic-lights) David's Model works great, it's the my first recommendation if you wanna try this code.
You can use Apple's coremltools to convert it to a MLModel, set the input/output interface you want, and integrate your model with my code.
TrafficLight Project provide full support for video capture, AVCapture arguments setting, and CoreML, Vision Integration, it's more about a sample or a toy，dont try it in realworld driving.