The quality of automatic recognition depends directly on the quality of the photos. Here are some best practices for each algorithm to get the best results.

---

Rules common to all algorithms #

Regardless of the method used, these principles apply systematically:

• Sharpness — avoid motion blur, especially on faces and bib numbers. Adjust your shutter speed to the sport you are photographing.
• Brightness — avoid backlighting and pronounced shadows on the face or bib
• Size in frame — the face or bib number must be large enough in the image to be detected. Avoid overly wide shots where the elements to be identified are tiny.
• Occlusion — avoid situations where the face or bib number is partially obscured by an arm, another runner, equipment, etc.

---

Face + Bib Number Algorithm #

• Choose front or three-quarter front angles to maximise the simultaneous visibility of the face and race number.
• Anticipate points where athletes naturally slow down (climbs, turns, finish lines) — the sharpness will be better.
• Ensure that the race number is clearly visible and not folded, creased or covered by clothing.
• Configure the min/max bib values in the album settings to limit false detections — see the article "Set restrictions on bib numbers".
• Organise a photocall at the start to enrich facial prints — see the article "Making the most of photocalls".

---

Algorithm Without a bib number #

• Photograph participants from the front or three-quarter front, with their face clearly visible.
• Avoid full-face helmets, balaclavas, sunglasses and any equipment that masks the face.
• The quality of recognition depends entirely on the face — pay particular attention to facial lighting.
• Organise a photocall beforehand to create high-quality reference facial prints — this is particularly important for this algorithm.
• For team sports, try to photograph each player individually at least once at the start of the match or training session.

---

Effective Algorithm #

• Update your team's reference photos at the start of the season — an old photo can impair recognition if the member's appearance has changed.
• The reference photograph must be: facing forward, clear, well lit, with an unobstructed face (no sunglasses, helmet or headwear).
• In real-life situations, photograph athletes from the front or three-quarter front view with their faces clearly visible.
• For sports involving protective equipment (helmets, balaclavas, etc.), please arrange for a photo session without equipment for reference photos.

---

Motorcycle Algorithm #

• Position yourself facing the motorbike or at a three-quarter angle to maximise the visibility of the race number on the fairing, helmet or handlebars.
• Avoid back or side angles where the number may be missing or illegible.
• Adjust your shutter speed to freeze motion — a blurred number will not be recognised.
• Configure word exclusions in the album settings to avoid false detections on number plates or advertisements — see the article "Motorcycle recognition".
• Focus on the slower sections of the course (chicanes, hairpin bends, technical sections) for sharper photos.

---

Car Algorithm #

• Take a photograph from the front or three-quarter front to maximise visibility of the number on the door, bonnet or roof.
• Slow-moving areas (departure, service area, hairpin bends) offer the best conditions for sharpness and legibility.
• If number plates generate false detections, add them to the exclusion words in the album settings — see the article "Car/rally recognition".
• For vehicles with numbers on multiple sides, vary your shooting angles to increase detection opportunities.

---

The photocall — a universal lever #

Regardless of the algorithm used, organising a photocall prior to the event significantly improves the quality of identifications:

• Choose a well-lit location, ideally with natural light.
• Opt for a neutral background that contrasts with the participants.
• Ask participants to stand facing forward, with their bibs clearly visible and unobstructed.
• This is the ideal moment to capture clear, front-facing faces before the effort begins.

See the article “Using photocalls” for strategies on how to use them according to your algorithm.

---

💡 One good photo is better than ten blurry ones. It is better to reduce the volume of imported photos by prioritising quality rather than overwhelming the algorithm with unusable images — this saves credits and improves identification accuracy.