Bosch 5100i IR Camera Review
Compact and discrete - big image shows monochrome performance in sub 2 lux.
Bosch 5100i IR Camera Review
♦ Bosch 5100i IR camera review – Bosch 5100i IR is an IP-66, IK-10 and NEMA 4X rated, cast alloy day/night camera offering 6MP of resolution (there’s a 12MP option) and all-round IR with a range of 20 metres.
There’s 1/1.8-inch CMOS sensor, the board-mount lens has a fixed focal length of 1.155mm with a fixed aperture of F2.0 giving a minimum focusing distance of 0.1 metre. The camera is NDAA-compliant, and resolution is 2112 x 2112 pixels with a total of 4.5MP used.
The camera delivers a big 360-degree view and there are viewing options that allow this to be displayed in a flexible and familiar way, including panoramic wall-mount 180 degrees, e-PTZ view, quad view, double panoramic and ceiling mount with corridor view.
Minimum scene illumination is 0.008 lux in monochrome with IR off, 0 lux when it’s activated, and 0.150 lux in colour. Video compression options include H.264, M-JPEG and H.265/HEVC, while WDR is 120dB.
There’s support for Micro SDXC/SDHC/ SD cards, as well as internal RAM pre-alarm recording of 5 seconds. The camera has an audio in and out, alarm in and out, video for HDMI, micro-HDMI, microphone integrated array and RJ-45 port. Audio streaming options are G.711 and AAC-LC, audio streaming is full or half-duplex, and signal-to-noise ratio is >50dB.
Data security is handled by secure element (TPM) and there’s RSA4096-bit, AES/CBC 256-bit PKI X.509 certificate encryption and end-to-end encryption with supported VMS via AES128 or AES256. Local storage encryption is XTS-AES and video authentication checksum options include MD5, SHA-1 and SHA-256. The camera delivers interoperability via ONVIF Profile S, ONVIF Profile G, ONVIF Profile M and ONVIF Profile T. There’s a raft of Bosch cyber security and networking functionality.
Camera housing colour is white, and the unit includes a gyro sensor, has a clear poly dome bubble with UV blocking and anti-scratch coating, while the cast aluminium housing has a dehumidifying membrane and a waterproof connection area.
This camera has an operating temperature range of -40 to 50C, very compact dimensions of 148 x 70mm, and it weighs 820g with the base plate. The camera uses PoE Type 1 Class 3 with a maximum consumption of 11.4W or 13.4W using 12V DC.
There’s a mounting plate included for surface mount, 4-inch square junction box, single and double gang box, and conduit compatibility supports a 3/4-inch back box.
Camera settings include contrast, saturation level and control, brightness, and sharpness, while white balance can be set from 2500 to 10,000K, and there are 3 automatic modes (basic, standard, sodium vapor), with a manual and a hold mode.
The camera has an automatic electronic shutter (AES), backlight compensation, dynamic noise reduction, 8 independent privacy masks, fully programmable scene modes, multiple default modes with scheduler, and display stamping with individual camera views and channels.
There’s a pixel counter, video watermarking, video content analysis, including intelligent video analytics, camera trainer, and alarm rules, including any object, object in field, line crossing, enter/leave field, loitering, follow route, idle/removed object, counting, occupancy, crowd density estimation, condition change, similarity search, flow/counter flow, while object filters can be set by duration, size, aspect ratio, speed, direction, colour, or by object classes. Tracking modes include standard (2D) tracking, 3D tracking, 3D people tracking, ship tracking and museum mode.
Built-in Bosch Intelligent Video Analytics and audio AI allow the camera to be set to trigger relevant alerts and quickly retrieve data, while the intelligent tracking allows movement to be followed continuously through the full image circle. Thanks to the fisheye lens, there’s no need to hand off tracking from one camera to another, simplifying movement analysis.
The system detects, tracks, and analyses objects, and alerts you when predefined alarms are triggered. A smart set of alarm rules makes complex tasks easier and reduces false alarms to a minimum. Moving objects can be classified in 4 different object classes: person, car, bike, truck. Meanwhile, audio AI gunshot detection, glass-break detection, and loud noises detection expand the camera’s ability to deliver situational awareness.
Calibration is quick and easy – as installer, you just enter the mounting height of the camera, and the internal gyro/accelerometer sensor provides the rest of the information to precisely calibrate video analytics.
Test Driving the Bosch 5100 IR
The Bosch 5100 IR panoramic camera is a robust external fisheye camera made of cast aluminium and poly – it’s compact enough for discretion, and while it feels solid, it’s not too heavy.
The 5100i IR proves easy to set up on SEN’s test network – my initial fire up of the camera is in the office and I note colour rendition is excellent, and detail strong, as is management of blooming from adjacent downlights. After an attempt to hang the camera over the back fence I abandon on the basis the potential of the scene is far too cluttered for a camera with this level of optical omniscience, my next port of call is the front of the office.
Having the camera mounted on the Magic Arm works well and a long pendant mount is probably the ideal setup as it gets the lens away from walls. You could pole mount the pendant – obviously it would be impossible to avoid having the pole appear in the angle of view, but it would only be a few degrees you’d lose.
Once I get the camera out front of the office on the Magic Arm, it’s clear this Bosch 5100i IR has a stupendous angle of view, giving staggering levels of situational awareness. The panoramic display covers the street from end to end, as well as encompassing the front of the office, the nearby buildings on this side of the street, and looking over to the 4th storey of the building opposite.
I have to decide which of the optional views is going to get me the best look at this busy inner city world and I decide to stick with the panoramic view, undertaking a double mouse click each time I want to get a closer look at the scene – it gives me a one-step digital zoom that allows the panorama to nearly fill the screen without too much scene shaved off in software or so much zoom I encounter pixel spread.
There are a bunch of display options for users in the real world, remember, including 180-degrees for video walls, double panorama, e-PTZ, quad view, and a ceiling mount with corridor view. No doubt these could be further enhanced using compatible VMS.
Because I’m looking outwards from the office at a street running at right angles, I quickly realise I’ve got more camera than view. This doesn’t pose a problem, because I can readily tilt the camera on the Magic Arm in order to centre the sweet spot of the lens while maximising coverage – an installer would need to choose tilt before installation.
When you’re looking at the images here it’s important to bear in mind that everything you can see from the front of the office to the other side of the street it would be possible to capture in the other direction as well – essentially, the camera is capable of viewing 2 streets, not just 1 – or 360 degrees, not 180.
Being a panoramic camera, there’s barrel distortion in the all-round image that’s most pronounced at the edges, just as you’d expect, but it’s not detracting from situational awareness. Pixel peeping, I note chromatic aberration – latitudinal and longitudinal. Maybe it’s 5 pixels deep. Given this combination of huge lens and modest sensor resolution, CAs are not likely to get in the way of your operational overview.
One of the nice things about all Bosch cameras is their intelligent analytics. I really like the orange outline placed around moving objects and the green line indicating direction of travel. Sitting at a workstation these indicators quickly become intuitive when you’re scanning an image – later I’m pleased to find it works just as well in low light.
IVA indicators are particularly useful in a scene where I’ve got an entire street on screen – there’s so much view I’m not unhappy to have automated assistance when it comes to keeping an eye on what’s happening. There’s a point early on that I adjust the camera to put the road and the footpath in the sweet spot in the middle of the camera view. But I still have more of this scene than we’ve ever had.
Our scene has got a lot of bright light on the far side, with deep shadow near the lens – it’s our typical daylight situation in the inner city. And the camera is doing well. I start with WDR off, turn it on and then later decide I prefer the unworked image.
With a view this large, modest sensor resolution of 2112 x 2112 pixels (4.5MP of the 6MP total viewable) means there will be some pixel spread and the question I wonder straight up is what impact this will have on depth of field. Bear in mind there is a 12MP version of this camera, but obviously those smaller photo-sites translate to a reduction in low light performance.
My instinct is that I’m not going to get face recognition but looking at people approaching and walking underneath the camera, it’s very close. I may not have DORI levels of face identification, but I certainly have recognition out towards the 10-metre mark and there’s good recognition of general details much further out – skin tones, hair colour, carried items, clothing colour.
As the afternoon goes along, I focus my attention on fast moving vehicles – because of this rectilinear lens, these are going past the sensor at near right angles – it’s not like a bullet camera with a motorised lens wound in to 6mm, which gives a tighter, nearer head-on field of view. The capacity of the Bosch camera to handle this sort of movement is strong and there’s very little motion blur. In one image I can see the attire of the driver in a vehicle going at least 40kmph, while adjacent depth of field remains excellent.
We’ve pointed this out before when testing panoramic cameras but it’s impossible not to highlight it again – you just don’t miss anything across a scene when it comes to situational awareness with the Bosch 5100 IR. I’m able to view some one approach the front door of the office, while workers sit outside the house directly opposite, a person goes into a diagonally opposite apartment building, a vehicle comes up the road at one end, another goes down the road at the other, and another vehicle fudges reverse parking 30 metres up on my side, and I count 4 separate pedestrian interactions – all in the 1 image.
As vehicles pass, I get a good sense that in reasonable light motion blur is low and I’ve got good detail of vehicles closer to the lens – within about 12 metres. I don’t get plates in this street view. When it comes to pedestrians, I’ve got face recognition inside 8 metres and directly under the camera at about 4 metres, I’ve got identification. Situational awareness remains massive. It’s all the way up and down the street, softening at the farther ends but still very worthwhile.
Next, I start scrolling around inside an image to see how much useful detail I can winkle out. The afternoon is going by, but the light is still good. I note good situational awareness on both sides of the sides of the street. A person walks past close to the lens. I’ve got tie colour, hair colour, the colour of shoes. On the other side there’s a bit of softness evident from pixel spread but I’ve still got clothing, shoe, and hair colour.
Looking deeper into the scene, at about 35 metres I’ve got a pedestrian walking away wearing blue jeans, blue top, beige backpack, dark hair, light-coloured shoes. At the same time, a man walks past at 7 metres wearing brown shoes, a purple shirt and a blue jacket, while over on the far side of the street a couple of workmen are sitting at the front of a terrace eating a snack. I double click on the image and step in to view a vehicle turning into the bottom end of the street.
As the sun moves in and out of the cloud cover impacting on shaded areas, I decide to nudge wide dynamic range. You expect WDR to put work on an image as the scene is universally brightened and that’s what happens here. The light levels across the scene are more uniform, stronger colours – oranges and blues – become more saturated and take on a subtle glow, and the overall contrast of the scene is reduced somewhat. Activating WDR is revealing in the shaded areas.
I look at the workers over the road in bright sunlight – they’re about 25 metres from the lens. I don’t have faces. Colour rendition is reduced a little with WDR activated, which is typical, in my opinion. I can see skin tone, orange top, brown shorts, hair colour, dark shoes or boots, and see the yellow high-vis on another chap. A few people are walking around elsewhere in the scene over there and I can see details like dark hair, puffer jacket, white shoes.
After spending a bit of time with WDR activated I can’t conclusively decide whether I prefer it on, WDR certainly opens up the dark areas of the scene when there’s bright light elsewhere. WDR is best used for extreme conditions – that’s my opinion.
When it comes to pixel spread in good light, I find that double click, which steps in and fills the 1920 x 1080p monitor with most the image, to be the ideal degree of zoom. If you go too far with digital zoom detail softens. This is not surprising – the 5100 IR 6MP is not about high resolution, it’s about exceptional situational awareness. Resolution is sufficient that when you wind in a little, you do get more detail. I should point out that not everyone is going to attempt to surveil 200 metres of city street with the 5100 IR – in more compact spaces, this camera is going to excel even more.
As the afternoon fades, I still have recognition of people inside the 12-metre zone, along with all sorts of fine detail, including shoe style and cut of clothing – pockets on jackets – the sorts of small things that can make a difference to an investigation when you’re trying to snare ID in failing light. When I go searching over the road – it’s 20 metres or more from the lens – I find I can’t get the same levels of detail but what I get is still worthwhile.
At this point, a group of people goes past on my side of the street, and at the same moment a vehicle goes past – there’s a man standing behind a 4WD vehicle waiting for traffic to pass so he can open his car door – there’s also there’s a person on the far side of the road walking up towards Albion St. There’s another person walking home from work, and I’ve got coat colour and style, hair colour, I can see he’s wearing a backpack and looking at his phone. Another very full image.
I try to find the distance the camera loses near face recognition in this failing light and it’s at around 8-10 metres, but I still have the same street-wide view of traffic and pedestrians I’m finding so pleasing. There’s no doubt that edging into the image using mouse wheel zoom causes things to get soft past the 20-metre mark, but this doesn’t change the value of the situational awareness deeper in, where I retain details of attire and appearance, as well as getting vehicle detail.
Low light Performance
I test the camera after dark – it’s sub 2 lux under the lens. I’m not sure what to expect from the 5100 IR 6MP in low light but as soon as I sit down at my workstation, I can see the camera is doing a great job scrounging reflectance from streetlamps and windows. Even though the sky is dark, it’s contriving to show blue-grey and throughout the scene there’s loads of detail. There’s some amplification noise in this image but whatever the camera engine is doing, it’s working.
In the first image I have a person at 16 metres in blue jeans, a dark grey top and white casual shoes with a navy duffle bag slung over his shoulder. He has a fair complexion and short brown hair. Across the road a woman is coming down the street. She has a fair complexion, dark hair pulled up, white shoes and socks, a light brown or purple dress, a beige coat and is carrying a handbag. I don’t have face identification, but what I have instead is this same level of detail street-wide.
Colour rendition of static vehicles is solid – blues, reds, greys, blacks, whites, creams, silvers – all these colours are easy to distinguish and opening the door of the office, I can see colour rendition is true. On the monitor I see the shutter has slowed down – it’s probably not at default 1/30th of a second. Looking at the balance of blur, amplification and light across this scene, I decide to let the camera continue managing on its own.
At sub 2 lux there’s motion blur directly under the lens but at the same time I can see hair and clothing colour – I’ve lost face recognition. A vehicle goes by – there’s a touch more light on the crown of the road – I can see the car is a charcoal Toyota Corolla sedan.
A couple walks down the path and I have plenty of clothing detail and I can see a guy further up the road looking at a phone – he’s in jeans, white shirt, dark jacket, light coloured shoes. In this image a car zooms by and it might be the one from earlier but going quicker, making it harder to be certain.
A chap comes down the footpath in a white shirt, black pants, black shoes and a chestnut leather jacket. With a little help from the light above a front door, I have face recognition. In the same scene back in the opposite direction around 30m from the lens, a woman is standing on the path in a white top and black pants. Someone else comes past and I’ve got observation levels of detail, including details of a tie.
Considering how dark it is out here, the camera is doing a great job in colour. I’ve lost none of that pleasing depth of field I had during the day. It gets soft deeper in, but this is about following scene-wide events – being able to see movement 40 metres from the lens at opposite ends of a street from a single camera point.
I note that Bosch’s analytics are managing to clearly detect areas of movement and reliably show direction of travel as well. There are times when a person in a dark spot is not immediately picked up. But there’s also an instance when the camera’s IVA snares a person way over the road towards the pub, opening the side door of a white van. That performance surprises me.
Going upstairs and looking down on the variably lit scene out the front of the office helps me to appreciate the camera’s uniform rendering when I go back down to the workstation. You wouldn’t think there’s only a few lux falling on this scene. In the brighter areas the camera does a little better, but it does well in the shadows, too. And the situational awareness is, well, staggering really.
I push the camera into night mode and IR comes on and my immediate reaction is that it’s a great spread – not only good depth but great lateral reach. The specification claims IR stretches 20 metres, but I think it’s pushing past 20 metres, perhaps reaching 25.
Going into monochrome has reduced motion blur a touch. When a cab zooms up the street, I can make out the shape of the sign writing on its door, as well as discerning considerable detail of the vehicle. A person walks up the street wearing a backpack – no problem with analytics picking up movement and direction of travel with vehicles and pedestrians in monochrome, either.
There’s a headlight down the far end of the street – about 80 metres from the lens – and its low angle provokes a moment of blooming and some internal reflections between sensor and dome. It’s probably a high beam as I don’t see this characteristic again despite plenty of cars going by. The nature of this camera test is a little unusual for the 5100 IR – mounting height is around 3 metres, and the camera is tilted upwards at the outer edge.
A person goes past on the far side of the road. I can see a white top dark hair, dark pants, I even get a sense of the way this person walks. Next, a couple of guys walk under the lens, and I can see what they’re wearing as well as details of haircuts and face shape. It’s not face ID but there’s plenty enough to give thoughtful investigators considered recognition throughout this epic scene.
In a classic Bosch 5100 IR scenario, a person parks their car and crosses the road out front, with IVA tracking them. A vehicle has pulled up further down the hill. A guy is walking up the hill towards the camera, behind him a couple is walking in the opposite direction across the lane towards Central Station, and in the other direction there’s a car parking. That’s really the strength of panoramic cameras – the ‘everything’ that you get in one view.
On impulse, I grab a static image of the street with nothing happening just because it’s sharp and flinty. Rolling the mouse, I find pixel spread and uncover a little amplification noise but that doesn’t impact on my satisfaction. A guy walks past underneath the camera, and you can see his backpack, including the cut of the straps and zipper placement. I go poking through this image looking for something else and while there’s no other activity, the fact I’m instinctively doing this tells you something.
Before ending the test, I switch back into colour – the transition confirms colour offers a strong image and I note the camera’s doing well on the opposite side of the road just on the edge of the sweet spot. When a pedestrian goes by over there, I’ve got either shoes or sock colour. Camouflage type pants, dark top, short haircut, the person is carrying a satchel or shoulder bag. In colour at night tones are strong and natural.
In the last image a guy walking is walking up the street towards the lens with a reversed cap, dark jacket, grey pants, and black boots. There’s a woman with fair hair standing on the street behind him, perhaps waiting for an Uber. Over the road at about 35 metres from the lens a person is walking down the hill. I note I’m not getting static plates, even those quite close to the lens, though I have no trouble with vehicle make, model and colour.
Investigators know that there are times when a case will hinge on the merest seconds of footage showing a person walking through a tight camera angle in poor light, often without face recognition and with ID hinging on characteristics like gait supported by triangulation provided by mobile phone data. Conversely, a camera like this gives investigators complete visual context.
The big thing with this camera is its massive angle of view – an angle of view that our 180-degree street scene doesn’t adequately reveal. I used the phrase ‘optical omniscience’ to describe the characteristics of the 5100 IR earlier, and this observation remains valid at the end of our test.
As well as offering coverage of everything in complex scenes to a reasonable depth, the 5100 IR would excel on building corners and when pendant-mounted in spaces demanding saturation coverage in all directions.
When you consider the situational awareness on offer, the balance of face recognition and ID to about 12 metres, and the attire – shoes, pants, shirts, hats, carried items, hair colour, skin tones – you can find deeper in, it’s impossible not to be satisfied. In the sweet spot between 6-8 metres, the camera goes very close to face identification during the day and delivers high levels of detail at night, especially in night mode.
Every single-sensor panoramic camera must balance resolution and angle of view. At the end of this test, I find I’m comfortable with that trade-off because while this lens costs me detail, it delivers me detail as well – vast detail.
In short, the Bosch 5100 IR is a situational awareness monster that delivers surprising colour in sub 2 lux, as well as performing very well in monochrome supported by its excellent IR. If you need to see everything from a single camera point at a reasonable price, and if you value a mature IVA suite, the Bosch 5100 IR is well worth a look.
#SEN #SENnews #security #electronics