Trench coats, umbrellas and bright sunlight can be the enemy of pedestrian detection.
Autonomous emergency braking (AEB) with pedestrian detection is one of the must-have safety features in modern cars, but it's not infallible. Whether it's false positives caused by reflective signs or systems failing during demonstrations, semi-autonomous safety assists can be fooled.
We're aware no system is perfect, as evidenced by this particularly public stuff-up, but a list published in Toyota's owner manual highlights the full list of scenarios where its pedestrian detection software might not actually detect a pedestrian. It's worth bearing in mind, these limitations aren't exclusive to Toyota, and similar lists exist for most brands.
The list of things it mightn't be able to recognise is below:
- Pedestrians shorter than 1m or taller than 2m
- Pedestrians wearing oversized clothing (raincoat), making their silhouette obscure
- Pedestrians carrying large baggage, holding an umbrella (etc.) hiding part of their body
- Pedestrians bending forward or squatting
- Groups of pedestrians who are extremely close together
- Pedestrians wearing white, who look extremely bright
- Pedestrians in the dark (eg. at night, or in a tunnel)
- Pedestrians whose clothing appears to be nearly the same colour/brightness as their surroundings
- Pedestrians near walls, fences, guardrails or large objects
- Pedestrians standing on a metal object like a manhole cover
- Pedestrians walking fast, or abruptly changing their speed
- Pedestrians walking out from behind a vehicle or large object
Exhaustive, right? The list was posted on Twitter by user Timothy B. Lee.
It's worth bearing in mind, these systems are only semi-autonomous, and don't claim to be perfect. Toyota relies on a radar sensor behind the badge, and a camera mounted behind the windscreen for the operation of its semi-autonomous assists, which include AEB, radar cruise control and lane-keeping assist.
Those sensors can be impacted by dust or road grime buildup, glare or complex visual scenes (people standing in front of odd-shaped objects, or holding things like umbrellas), potentially undermining their effectiveness. Even Tesla, which has painted itself as a leader in the autonomous field with Autopilot, has previously struggled with object detection.
“Autonomous [technology] is not perfect yet... especially the city area, there are many unexpected situations happening," he said.
"Autonomous [driving] in city is a very dangerous situation sometimes, but freeway or toll road things are quite different. So I think first of all we focus on the freeway and highway to satisfy the autonomous [demands], but then later on we are going to introduce autonomous for city area," he continued.
"But before [that] we focus on safety to reduce the collision or accident or injury. Some may say that’s conservative but our priority is more on the safety issue.”
MORE: Autonomous driving news