What I ended up doing was making a prefab which contains a box collider and script. On Update, check if ray cast from camera sees it, hits it. I figured this way, I'd be testing against a simple shape, it wouldn't affect my model prefab (sometimes need to reimport or swap them out), and I just put these around objects I care about. Seems to work pretty well.

You would probably want to use something thicker than a Raycast such as a capsule cast, if you are going to use some sort of a cast.

Another easy way would just be to use renderer.isVisible.

Renderer[] renderers = (Renderer[])(FindObjectsOfType(typeof(Renderer)));
//You probably want to cache this at the beginning.

foreach(Renderer r in renderers) {
      if(r.isVisible) {
            //Do something.
      }
}

Or, you could use the Dot product of the camera's forward Vector and the difference between the object's position and the camera's position. This would probably overly complicated, but Vector math is inexpensive for the most part, and if you can't use a the renderer method, then this might be a good idea. Personally, this is probably more complicated than you want to make it, but it's just another (albeit a bit strange) option.

Transform[] sceneObjects = (Transform[])(FindObjectsOfType(typeof(Transform)));

Vector3 thisObjPos = transform.position;
Vector3 forward = transform.forward;
//finding them now will save us from finding them when we iterate through.

foreach(Transform sceneObject : Transform in sceneObjects) {
      Vector3 offset = sceneObject.position - thisObjPos;
      offset = offset.normalized;

      If(Vector3.Dot(forward, offset) > someValue /*between -1, 1*/) {
           //Do stuff
      }
}

Check if a screen rect contains the object's position using WorldToScreenPoint:

var percentageOfScreenHeight = .25;
private var centerRect : Rect;

function Start () {
    var ySize = Screen.height*percentageOfScreenHeight;
    centerRect = Rect(Screen.width/2 - ySize/2, Screen.height/2 - ySize/2, ySize, ySize);
}

function Update () {
    if (centerRect.Contains(Camera.main.WorldToScreenPoint(transform.position))) {
        Debug.Log ("Looked at");
    }
}

Since you only need it twice per second, you'd use InvokeRepeating instead of Update. You'd also want to make sure the objects are in front of the camera; the most optimal thing would probably be to make a List of transforms, and have the objects add/remove themselves to and from the list using OnBecameVisible and OnBecameInvisible, then cycle through that array (instead of having each object run that script). You can use Vector3.Distance to see if an object that's being looked at is close enough.