Support Center

Tips for getting best recognition

Jun 21, 2017
  1. Where you place your camera can dramatically affect your recognition accuracy. If you can place the camera in a well-lit area where you can see the entire person, your results should be very good. The following are categories of problems that can affect recognition:
    • Poor light. If you want to use the system at night, it won't recognize people outside of well-lit areas. Leaving a porch light on, for example, will increase accuracy.
    • Dark shadows and low contrast. If it's possible, avoid areas where a person will blend into the background. For example, people tend to stand out more clearly against a light-colored wall than they do under shady trees.
    • Parts of people covered up.
    • When people pass behind objects like trees or table, they appear to be split into smaller objects, which can be difficult to detect.  
    • Accuracy is best if the full figure is visible for as long as possible. For example, the camera may be at an angle where people passing through are partially cut off through most of the clip. Repositioning the camera can eliminate many of these problems.  
    • If people are too close to the camera, and large parts are out of the frame, it may be difficult to detect them accurately  
    • Trees and branches. Swaying branches and shifting leaves in bushes can be picked up as objects or even people. Sometimes pointing the camera towards the lawn can crop out problem parts. A branch near the camera may be detected as an object if the hard edges of the leaves sway in the wind.  
    • Reflections. Windows or water surfaces near streets can reflect headlights that may appear to be objects. If you have a camera inside a window pointing out, you may get reflections from the inside. Try to place the lens flush against the glass if possible.  
    • Overhead angles. The software has not been trained with overhead views, so pointing a camera straight down on someone will not generate good results.
  2. Frame rate is important. If the motion in your videos appears jerky, the frame rate is probably under 10 frames per second. See "The video is jerky (low frame rate)" for more details.
  3. Camera resolution. IP cameras typically can be set to display video at different resolutions. The minimum recommended resolution is 320 x 240 (QVGA). 640 x 480 resolution should not in theory give you better results, since the application scales video down to QVGA before processing. If the quality of your camera's QVGA is poor, however, setting it to VGA may be slightly better. On the other hand, if you have good recognition at 320 x 240, it can be beneficial to use the lower resolution. Lower resolution means less data that needs to be sent across your wireless network and processed by your computer (in case your computer seems sluggish when you are running the Sighthound Video application). 
  4. Contrast is important. Webcams are typically optimized for indoor use, and can appear washed out when pointed outside. You may need to adjust the exposure on the camera, which is described above in the section "My webcam video is washed out."

Tip: if you want to be conservative, you can try looking for all objects. The person may have been detected as a moving object, but not recognized as a person.